Transcript
00:00:00 The following is a conversation with Jay McClelland,
00:00:03 a cognitive scientist at Stanford
00:00:05 and one of the seminal figures
00:00:06 in the history of artificial intelligence
00:00:09 and specifically neural networks.
00:00:12 Having written the parallel distributed processing book
00:00:15 with David Rommelhart,
00:00:17 who coauthored the backpropagation paper with Jeff Hinton.
00:00:21 In their collaborations, they’ve paved the way
00:00:24 for many of the ideas
00:00:25 at the center of the neural network based
00:00:27 machine learning revolution of the past 15 years.
00:00:32 To support this podcast,
00:00:33 please check out our sponsors in the description.
00:00:36 This is the Lex Friedman podcast
00:00:38 and here is my conversation with Jay McClelland.
00:00:43 You are one of the seminal figures
00:00:45 in the history of neural networks.
00:00:47 At the intersection of cognitive psychology
00:00:49 and computer science,
00:00:51 what to you has over the decades emerged
00:00:54 as the most beautiful aspect about neural networks?
00:00:57 Both artificial and biological.
00:01:00 The fundamental thing I think about with neural networks
00:01:03 is how they allow us to link
00:01:08 biology with the mysteries of thought.
00:01:17 When I was first entering the field myself
00:01:19 in the late 60s, early 70s,
00:01:23 cognitive psychology had just become a field.
00:01:29 There was a book published in 67 called Cognitive Psychology.
00:01:36 And the author said that the study of the nervous system
00:01:42 was only of peripheral interest.
00:01:44 It wasn’t going to tell us anything about the mind.
00:01:48 And I didn’t agree with that.
00:01:51 I always felt, oh, look, I’m a physical being.
00:01:58 From dust to dust, you know,
00:02:01 ashes to ashes, and somehow I emerged from that.
00:02:06 So that’s really interesting.
00:02:08 So there was a sense with cognitive psychology
00:02:11 that in understanding the neuronal structure of things,
00:02:17 you’re not going to be able to understand the mind.
00:02:20 And then your sense is if we study these neural networks,
00:02:23 we might be able to get at least very close
00:02:25 to understanding the fundamentals of the human mind.
00:02:28 Yeah.
00:02:29 I used to think, or I used to talk about the idea
00:02:32 of awakening from the Cartesian dream.
00:02:36 So Descartes, you know, thought about these things, right?
00:02:41 He was walking in the gardens of Versailles one day,
00:02:46 and he stepped on a stone.
00:02:48 And a statue moved.
00:02:52 And he walked a little further,
00:02:53 he stepped on another stone, and another statue moved.
00:02:55 And he, like, why did the statue move
00:02:59 when I stepped on the stone?
00:03:00 And he went and talked to the gardeners,
00:03:02 and he found out that they had a hydraulic system
00:03:06 that allowed the physical contact with the stone
00:03:10 to cause water to flow in various directions,
00:03:12 which caused water to flow into the statue
00:03:14 and move the statue.
00:03:15 And he used this as the beginnings of a theory
00:03:22 about how animals act.
00:03:28 And he had this notion that these little fibers
00:03:33 that people had identified that weren’t carrying the blood,
00:03:37 you know, were these little hydraulic tubes
00:03:39 that if you touch something, there would be pressure,
00:03:42 and it would send a signal of pressure
00:03:43 to the other parts of the system,
00:03:46 and that would cause action.
00:03:49 So he had a mechanistic theory of animal behavior.
00:03:54 And he thought that the human had this animal body,
00:04:00 but that some divine something else
00:04:03 had to have come down and been placed in him
00:04:06 to give him the ability to think, right?
00:04:10 So the physical world includes the body in action,
00:04:15 but it doesn’t include thought according to Descartes, right?
00:04:19 And so the study of physiology at that time
00:04:22 was the study of sensory systems and motor systems
00:04:26 and things that you could directly measure
00:04:30 when you stimulated neurons and stuff like that.
00:04:33 And the study of cognition was something that, you know,
00:04:38 was tied in with abstract computer algorithms
00:04:41 and things like that.
00:04:43 But when I was an undergraduate,
00:04:45 I learned about the physiological mechanisms.
00:04:48 And so when I’m studying cognitive psychology
00:04:51 as a first year PhD student, I’m saying,
00:04:53 wait a minute, the whole thing is biological, right?
00:04:56 You know?
00:04:57 You had that intuition right away.
00:04:59 That always seemed obvious to you.
00:05:00 Yeah, yeah.
00:05:03 Isn’t that magical, though,
00:05:04 that from just a little bit of biology can emerge
00:05:08 the full beauty of the human experience?
00:05:10 Why is that so obvious to you?
00:05:13 Well, obvious and not obvious at the same time.
00:05:18 And I think about Darwin in this context, too,
00:05:20 because Darwin knew very early on
00:05:25 that none of the ideas that anybody had ever offered
00:05:29 gave him a sense of understanding
00:05:31 how evolution could have worked.
00:05:36 But he wanted to figure out how it could have worked.
00:05:40 That was his goal.
00:05:42 And he spent a lot of time working on this idea
00:05:48 and reading about things that gave him hints
00:05:52 and thinking they were interesting but not knowing why
00:05:54 and drawing more and more pictures of different birds
00:05:57 that differ slightly from each other and so on, you know.
00:06:00 And then he figured it out.
00:06:03 But after he figured it out, he had nightmares about it.
00:06:06 He would dream about the complexity of the eye
00:06:10 and the arguments that people had given
00:06:12 about how ridiculous it was to imagine
00:06:16 that that could have ever emerged
00:06:19 from some sort of, you know, unguided process, right?
00:06:24 That it hadn’t been the product of design.
00:06:28 And so he didn’t publish for a long time,
00:06:32 in part because he was scared of his own ideas.
00:06:35 He didn’t think they could possibly be true.
00:06:40 But then, you know, by the time
00:06:44 the 20th century rolls around, we all,
00:06:49 you know, we understand that,
00:06:52 many people understand or believe
00:06:55 that evolution produced, you know, the entire
00:06:59 range of animals that there are.
00:07:03 And, you know, Descartes’s idea starts to seem
00:07:06 a little wonky after a while, right?
00:07:08 Like, well, wait a minute.
00:07:11 There’s the apes and the chimpanzees and the bonobos
00:07:15 and, you know, like, they’re pretty smart in some ways.
00:07:18 You know, so what?
00:07:20 Oh, you know, somebody comes up,
00:07:22 oh, there’s a certain part of the brain
00:07:23 that’s still different.
00:07:24 They don’t, you know, there’s no hippocampus
00:07:26 in the monkey brain.
00:07:28 It’s only in the human brain.
00:07:31 Huxley had to do a surgery in front of many, many people
00:07:34 in the late 19th century to show to them
00:07:36 there’s actually a hippocampus in the chimpanzee’s brain.
00:07:40 You know, so the continuity of the species
00:07:45 is another element that, you know,
00:07:49 contributes to this sort of, you know, idea
00:07:56 that we are ourselves a total product of nature.
00:08:01 And that, to me, is the magic and the mystery,
00:08:06 how nature could actually, you know,
00:08:11 give rise to organisms that have the capabilities
00:08:16 that we have.
00:08:20 So it’s interesting because even the idea of evolution
00:08:23 is hard for me to keep all together in my mind.
00:08:27 So because we think of a human time scale,
00:08:30 it’s hard to imagine, like, the development
00:08:33 of the human eye would give me nightmares too.
00:08:36 Because you have to think across many, many, many
00:08:38 generations, and it’s very tempting to think about
00:08:41 kind of a growth of a complicated object
00:08:44 and it’s like, how is it possible for such a thing
00:08:49 to be built?
00:08:50 Because also, me, from a robotics engineering perspective,
00:08:53 it’s very hard to build these systems.
00:08:55 How can, through an undirected process,
00:08:58 can a complex thing be designed?
00:09:00 It seems not, it seems wrong.
00:09:03 Yeah, so that’s absolutely right.
00:09:05 And I, you know, a slightly different career path
00:09:08 that would have been equally interesting to me
00:09:10 would have been to actually study the process
00:09:15 of embryological development flowing on
00:09:21 into brain development and the exquisite sort of laying
00:09:29 down of pathways and so on that occurs in the brain.
00:09:32 And I know the slightest bit about that is not my field,
00:09:35 but there are, you know, fascinating aspects
00:09:43 to this process that eventually result in the, you know,
00:09:49 the complexity of various brains.
00:09:54 At least, you know, one thing we’re,
00:09:59 in the field, I think people have felt for a long time,
00:10:02 in the study of vision, the continuity between humans
00:10:07 and nonhuman animals has been second nature
00:10:11 for a lot longer.
00:10:12 I was having, I had this conversation with somebody
00:10:16 who is a vision scientist and he was saying,
00:10:17 oh, we don’t have any problem with this.
00:10:19 You know, the monkey’s visual system
00:10:21 and the human visual system, extremely similar
00:10:26 up to certain levels, of course, they diverge after a while.
00:10:29 But the first, the visual pathway from the eye
00:10:34 to the brain and the first few layers of cortex
00:10:41 or cortical areas, I guess one would say,
00:10:45 are extremely similar.
00:10:49 Yeah, so on the cognition side is where the leap
00:10:52 seems to happen with humans,
00:10:54 that it does seem we’re kind of special.
00:10:56 And that’s a really interesting question
00:10:58 when thinking about alien life
00:11:00 or if there’s other intelligent alien civilizations
00:11:03 out there, is how special is this leap?
00:11:06 So one special thing seems to be the origin of life itself.
00:11:09 However you define that, there’s a gray area.
00:11:11 And the other leap, this is very biased perspective
00:11:14 of a human, is the origin of intelligence.
00:11:19 And again, from an engineer perspective,
00:11:22 it’s a difficult question to ask.
00:11:24 An important one is how difficult is that leap?
00:11:27 How special were humans?
00:11:30 Did a monolith come down?
00:11:32 Did aliens bring down a monolith
00:11:33 and some apes had to touch a monolith to get it?
00:11:38 That’s a lot like Descartes idea, right?
00:11:41 Exactly, but it just seems one heck of a leap
00:11:46 to get to this level of intelligence.
00:11:48 Yeah, and so Chomsky argued that some genetic fluke occurred
00:12:00 100,000 years ago and just happened
00:12:04 that some human, some hominin predecessor of current humans
00:12:13 had this one genetic tweak that resulted in language.
00:12:20 And language then provided this special thing that separates us
00:12:29 from all other animals.
00:12:36 I think there’s a lot of truth to the value and importance
00:12:39 of language, but I think it comes along
00:12:43 with the evolution of a lot of other related things related
00:12:48 to sociality and mutual engagement with others
00:12:53 and establishment of, I don’t know,
00:13:01 rich mechanisms for organizing and understanding
00:13:07 of the world, which language then plugs into.
00:13:12 Right, so language is a tool that
00:13:16 allows you to do this kind of collective intelligence.
00:13:18 And whatever is at the core of the thing that
00:13:21 allows for this collective intelligence is the main thing.
00:13:25 And it’s interesting to think about that one fluke, one
00:13:29 mutation could lead to the first crack opening of the door
00:13:36 to human intelligence.
00:13:38 All it takes is one.
00:13:39 Evolution just kind of opens the door a little bit,
00:13:41 and then time and selection takes care of the rest.
00:13:45 You know, there’s so many fascinating aspects
00:13:48 to these kinds of things.
00:13:49 So we think of evolution as continuous, right?
00:13:54 We think, oh, yes, OK, over 500 million years,
00:13:58 there could have been this relatively continuous changes.
00:14:04 And but that’s not what anthropologists,
00:14:12 evolutionary biologists found from the fossil record.
00:14:15 They found hundreds of millions of years of stasis.
00:14:24 And then suddenly a change occurs.
00:14:27 Well, suddenly on that scale is a million years or something,
00:14:32 or even 10 million years.
00:14:33 But the concept of punctuated equilibrium
00:14:38 was a very important concept in evolutionary biology.
00:14:44 And that also feels somehow right about the stages
00:14:53 of our mental abilities.
00:14:55 We seem to have a certain kind of mindset at a certain age.
00:14:59 And then at another age, we look at that four year old
00:15:04 and say, oh, my god, how could they have thought that way?
00:15:07 So Piaget was known for this kind of stage theory
00:15:10 of child development, right?
00:15:11 And you look at it closely, and suddenly those stages
00:15:14 are so discreet and it transitions.
00:15:17 But the difference between the four year old and the seven
00:15:19 year old is profound.
00:15:20 And that’s another thing that’s always interested me
00:15:24 is how something happens over the course of several years
00:15:29 of experience where at some point
00:15:31 we reach the point where something
00:15:33 like an insight or a transition or a new stage of development
00:15:37 occurs.
00:15:38 And these kinds of things can be understood
00:15:45 in complex systems research.
00:15:47 And so evolutionary biology, developmental biology,
00:15:55 cognitive development are all things
00:15:57 that have been approached in this kind of way.
00:15:59 Yeah.
00:16:01 Just like you said, I find both fascinating
00:16:03 those early years of human life, but also
00:16:07 the early minutes, days from the embryonic development
00:16:13 to how from embryos you get the brain.
00:16:17 That development, again, from an engineer perspective,
00:16:20 is fascinating.
00:16:22 So it’s not.
00:16:22 So the early, when you deploy the brain to the human world
00:16:27 and it gets to explore that world and learn,
00:16:29 that’s fascinating.
00:16:30 But just like the assembly of the mechanism
00:16:33 that is capable of learning, that’s amazing.
00:16:36 The stuff they’re doing with brain organoids
00:16:39 where you can build many brains and study
00:16:42 that self assembly of a mechanism from the DNA material,
00:16:48 that’s like, what the heck?
00:16:51 You have literally biological programs
00:16:55 that just generate a system, this mushy thing that’s
00:17:00 able to be robust and learn in a very unpredictable world
00:17:05 and learn seemingly arbitrary things,
00:17:08 or a very large number of things that enable survival.
00:17:14 Yeah.
00:17:15 Ultimately, that is a very important part
00:17:19 of the whole process of understanding
00:17:22 this emergence of mind from brain kind of thing.
00:17:27 And the whole thing seems to be pretty continuous.
00:17:29 So let me step back to neural networks
00:17:32 for another brief minute.
00:17:35 You wrote parallel distributed processing books
00:17:37 that explored ideas of neural networks in the 1980s
00:17:42 together with a few folks.
00:17:43 But the books you wrote with David Romelhart,
00:17:47 who is the first author on the back propagation
00:17:50 paper with Jeff Hinton.
00:17:52 So these are just some figures at the time
00:17:54 that we’re thinking about these big ideas.
00:17:57 What are some memorable moments of discovery
00:18:00 and beautiful ideas from those early days?
00:18:04 I’m going to start sort of with my own process in the mid 70s
00:18:13 and then into the late 70s when I met Jeff Hinton
00:18:18 and he came to San Diego and we were all together.
00:18:25 In my time in graduate schools, I’ve already described to you,
00:18:30 I had this sort of feeling of, OK, I’m
00:18:33 really interested in human cognition,
00:18:35 but this disembodied sort of way of thinking about it
00:18:40 that I’m getting from the current mode of thought about it
00:18:44 isn’t working fully for me.
00:18:47 And when I got my assistant professorship,
00:18:52 I went to UCSD and that was in 1974.
00:18:58 Something amazing had just happened.
00:19:00 Dave Romelhart had written a book together
00:19:03 with another man named Don Norman
00:19:06 and the book was called Explorations in Cognition.
00:19:09 And it was a series of chapters exploring
00:19:14 interesting questions about cognition,
00:19:17 but in a completely sort of abstract, nonbiological kind
00:19:22 of way.
00:19:23 And I’m saying, gee, this is amazing.
00:19:25 I’m coming to this community where people can get together
00:19:28 and feel like they’ve collectively exploring ideas.
00:19:35 And it was a book that had a lot of, I don’t know,
00:19:39 lightness to it.
00:19:40 And Don Norman, who was the more senior figure
00:19:47 to Romelhart at that time who led that project,
00:19:51 always created this spirit of playful exploration of ideas.
00:19:55 And so I’m like, wow, this is great.
00:19:58 But I was also still trying to get from the neurons
00:20:07 to the cognition.
00:20:10 And I realized at one point, I got this opportunity
00:20:15 to go to a conference where I heard a talk by a man named
00:20:18 James Anderson, who was an engineer,
00:20:22 but by then a professor in a psychology department, who
00:20:26 had used linear algebra to create neural network
00:20:32 models of perception and categorization and memory.
00:20:37 And it just blew me out of the water
00:20:41 that one could create a model that was simulating neurons,
00:20:47 not just engaged in a stepwise algorithmic process that
00:20:56 was construed abstractly.
00:20:58 But it was simulating remembering and recalling
00:21:03 and recognizing the prior occurrence of a stimulus
00:21:07 or something like that.
00:21:08 So for me, this was a bridge between the mind and the brain.
00:21:14 And I remember I was walking across campus one day in 1977,
00:21:20 and I almost felt like St. Paul on the road to Damascus.
00:21:25 I said to myself, if I think about the mind in terms
00:21:30 of a neural network, it will help
00:21:32 me answer the questions about the mind
00:21:33 that I’m trying to answer.
00:21:36 And that really excited me.
00:21:38 So I think that a lot of people were
00:21:43 becoming excited about that.
00:21:45 And one of those people was Jim Anderson, who I had mentioned.
00:21:49 Another one was Steve Grossberg, who
00:21:52 had been writing about neural networks since the 60s.
00:21:58 And Jeff Hinton was yet another.
00:22:00 And his PhD dissertation showed up in an applicant pool
00:22:08 to a postdoctoral training program
00:22:11 that Dave and Don, the two men I mentioned before,
00:22:16 Rumelhart and Norman, were administering.
00:22:19 And Rumelhart got really excited about Hinton’s PhD dissertation.
00:22:26 And so Hinton was one of the first people
00:22:30 who came and joined this group of postdoctoral scholars
00:22:34 that was funded by this wonderful grant that they got.
00:22:39 Another one who is also well known
00:22:41 in neural network circles is Paul Smolenski.
00:22:45 He was another one of that group.
00:22:47 Anyway, Jeff and Jim Anderson organized a conference
00:22:55 at UCSD where we were.
00:22:59 And it was called Parallel Models of Associative Memory.
00:23:04 And it brought all the people together
00:23:06 who had been thinking about these kinds of ideas
00:23:08 in 1979 or 1980.
00:23:11 And this began to kind of really resonate
00:23:18 with some of Rumelhart’s own thinking,
00:23:23 some of his reasons for wanting something
00:23:26 other than the kinds of computation
00:23:28 he’d been doing so far.
00:23:29 So let me talk about Rumelhart now for a minute,
00:23:32 OK, with that context.
00:23:33 Well, let me also just pause because he
00:23:34 said so many interesting things before we go to Rumelhart.
00:23:37 So first of all, for people who are not familiar,
00:23:40 neural networks are at the core of the machine learning,
00:23:43 deep learning revolution of today.
00:23:45 Geoffrey Hinton that we mentioned
00:23:46 is one of the figures that were important in the history
00:23:50 like yourself in the development of these neural networks,
00:23:53 artificial neural networks that are then
00:23:54 used for the machine learning application.
00:23:56 Like I mentioned, the backpropagation paper
00:23:59 is one of the optimization mechanisms
00:24:02 by which these networks can learn.
00:24:05 And the word parallel is really interesting.
00:24:09 So it’s almost like synonymous from a computational
00:24:12 perspective how you thought at the time about neural networks
00:24:17 as parallel computation.
00:24:20 Would that be fair to say?
00:24:21 Well, yeah, the parallel, the word parallel in this
00:24:25 comes from the idea that each neuron is
00:24:30 an independent computational unit, right?
00:24:33 It gathers data from other neurons,
00:24:36 it integrates it in a certain way,
00:24:39 and then it produces a result. And it’s
00:24:41 a very simple little computational unit.
00:24:44 But it’s autonomous in the sense that it does its thing, right?
00:24:51 It’s in a biological medium where
00:24:53 it’s getting nutrients and various chemicals
00:24:57 from that medium.
00:25:00 But you can think of it as almost like a little computer
00:25:05 in and of itself.
00:25:08 So the idea is that each our brains have, oh, look,
00:25:13 100 or hundreds, almost a billion
00:25:17 of these little neurons, right?
00:25:21 And they’re all capable of doing their work at the same time.
00:25:25 So it’s like instead of just a single central processor that’s
00:25:30 engaged in chug one step after another,
00:25:36 we have a billion of these little computational units
00:25:41 working at the same time.
00:25:42 So at the time that’s, I don’t know, maybe you can comment,
00:25:45 it seems to me, even still to me,
00:25:49 quite a revolutionary way to think about computation
00:25:52 relative to the development of theoretical computer science
00:25:56 alongside of that where it’s very much like sequential computer.
00:26:00 You’re analyzing algorithms that are running on a single computer.
00:26:04 You’re saying, wait a minute, why don’t we
00:26:08 take a really dumb, very simple computer
00:26:11 and just have a lot of them interconnected together?
00:26:14 And they’re all operating in their own little world
00:26:16 and they’re communicating with each other
00:26:18 and thinking of computation that way.
00:26:21 And from that kind of computation,
00:26:24 trying to understand how things like certain characteristics
00:26:28 of the human mind can emerge.
00:26:31 That’s quite a revolutionary way of thinking, I would say.
00:26:35 Well, yes, I agree with you.
00:26:37 And there’s still this sort of sense
00:26:44 of not sort of knowing how we kind of get all the way there,
00:26:53 I think.
00:26:54 And this very much remains at the core of the questions
00:26:58 that everybody’s asking about the capabilities
00:27:01 of deep learning and all these kinds of things.
00:27:02 But if I could just play this out a little bit,
00:27:07 a convolutional neural network or a CNN,
00:27:11 which many people may have heard of, is a set of,
00:27:19 you could think of it biologically as a set of
00:27:24 collections of neurons.
00:27:27 Each collection has maybe 10,000 neurons in it.
00:27:33 But there’s many layers.
00:27:35 Some of these things are hundreds or even
00:27:38 1,000 layers deep.
00:27:39 But others are closer to the biological brain
00:27:43 and maybe they’re like 20 layers deep or something like that.
00:27:47 So within each layer, we have thousands of neurons
00:27:52 or tens of thousands maybe.
00:27:54 Well, in the brain, we probably have millions in each layer.
00:27:59 But we’re getting sort of similar in a certain way.
00:28:05 And then we think, OK, at the bottom level,
00:28:09 there’s an array of things that are like the photoreceptors.
00:28:12 In the eye, they respond to the amount
00:28:14 of light of a certain wavelength at a certain location
00:28:17 on the pixel array.
00:28:21 So that’s like the biological eye.
00:28:24 And then there’s several further stages going up,
00:28:27 layers of these neuron like units.
00:28:30 And you go from that raw input array of pixels
00:28:36 to the classification, you’ve actually
00:28:40 built a system that could do the same kind of thing
00:28:44 that you and I do when we open our eyes and we look around
00:28:46 and we see there’s a cup, there’s a cell phone,
00:28:49 there’s a water bottle.
00:28:52 And these systems are doing that now, right?
00:28:54 So they are, in terms of the parallel idea
00:29:00 that we were talking about before,
00:29:02 they are doing this massively parallel computation
00:29:05 in the sense that each of the neurons in each
00:29:08 of those layers is thought of as computing
00:29:12 its little bit of something about the input
00:29:17 simultaneously with all the other ones in the same layer.
00:29:21 We get to the point of abstracting that away
00:29:24 and thinking, oh, it’s just one whole vector that’s
00:29:27 being computed, one activation pattern that’s
00:29:30 computed in a single step.
00:29:32 And that abstraction is useful, but it’s still that parallel.
00:29:39 And distributed processing, right?
00:29:41 Each one of these guys is just contributing
00:29:43 a tiny bit to that whole thing.
00:29:45 And that’s the excitement that you felt,
00:29:46 that from these simple things, you can emerge.
00:29:50 When you add these level of abstractions on it,
00:29:53 you can start getting all the beautiful things
00:29:56 that we think about as cognition.
00:29:58 And so, OK, so you have this conference.
00:30:01 I forgot the name already, but it’s
00:30:02 Parallel and Something Associative Memory and so on.
00:30:05 Very exciting, technical and exciting title.
00:30:08 And you started talking about Dave Romerhart.
00:30:11 So who is this person that was so,
00:30:15 you’ve spoken very highly of him.
00:30:17 Can you tell me about him, his ideas, his mind, who he was
00:30:22 as a human being, as a scientist?
00:30:24 So Dave came from a little tiny town in Western South Dakota.
00:30:31 And his mother was the librarian,
00:30:35 and his father was the editor of the newspaper.
00:30:41 And I know one of his brothers pretty well.
00:30:46 They grew up, there were four brothers,
00:30:49 and they grew up together.
00:30:53 And their father encouraged them to compete with each other
00:30:56 a lot.
00:30:58 They competed in sports, and they competed in mind games.
00:31:04 I don’t know, things like Sudoku and chess and various things
00:31:07 like that.
00:31:08 And Dave was a standout undergraduate.
00:31:16 He went at a younger age than most people
00:31:20 do to college at the University of South Dakota
00:31:23 and majored in mathematics.
00:31:24 And I don’t know how he got interested in psychology,
00:31:30 but he applied to the mathematical psychology
00:31:33 program at Stanford and was accepted as a PhD student
00:31:37 to study mathematical psychology at Stanford.
00:31:40 So mathematical psychology is the use of mathematics
00:31:46 to model mental processes.
00:31:50 So something that I think these days
00:31:52 might be called cognitive modeling, that whole space.
00:31:55 Yeah, it’s mathematical in the sense
00:31:57 that you say, if this is true and that is true,
00:32:05 then I can derive that this should follow.
00:32:08 And so you say, these are my stipulations
00:32:10 about the fundamental principles,
00:32:12 and this is my prediction about behavior.
00:32:15 And it’s all done with equations.
00:32:16 It’s not done with a computer simulation.
00:32:19 So you solve the equation, and that tells you
00:32:23 what the probability that the subject
00:32:26 will be correct on the seventh trial or the experiment is
00:32:29 or something like that.
00:32:30 So it’s a use of mathematics to descriptively characterize
00:32:37 aspects of behavior.
00:32:39 And Stanford at that time was the place
00:32:43 where there were several really, really strong
00:32:48 mathematical thinkers who were also connected with three
00:32:51 or four others around the country who brought
00:32:55 a lot of really exciting ideas onto the table.
00:32:59 And it was a very, very prestigious part
00:33:02 of the field of psychology at that time.
00:33:05 So Rummelhart comes into this.
00:33:08 He was a very strong student within that program.
00:33:13 And he got this job at this brand new university
00:33:19 in San Diego in 1967, where he’s one of the first assistant
00:33:24 professors in the Department of Psychology at UCSD.
00:33:30 So I got there in 74, seven years later,
00:33:37 and Rummelhart at that time was still
00:33:43 doing mathematical modeling.
00:33:48 But he had gotten interested in cognition.
00:33:53 He’d gotten interested in understanding.
00:33:58 And understanding, I think, remains,
00:34:04 what does it mean to understand anyway?
00:34:08 It’s an interesting sort of curious,
00:34:11 how would we know if we really understood something?
00:34:14 But he was interested in building machines
00:34:18 that would hear a couple of sentences
00:34:21 and have an insight about what was going on.
00:34:23 So for example, one of his favorite things at that time
00:34:26 was, Margie was sitting on the front step
00:34:32 when she heard the familiar jingle of the good humor man.
00:34:38 She remembered her birthday money and ran into the house.
00:34:42 What is Margie doing?
00:34:44 Why?
00:34:47 Well, there’s a couple of ideas you could have,
00:34:50 but the most natural one is that the good humor
00:34:53 man brings ice cream.
00:34:55 She likes ice cream.
00:34:57 She knows she needs money to buy ice cream,
00:34:59 so she’s going to run into the house and get her money
00:35:02 so she can buy herself an ice cream.
00:35:03 It’s a huge amount of inference that
00:35:05 has to happen to get those things to link up
00:35:07 with each other.
00:35:09 And he was interested in how the hell that could happen.
00:35:13 And he was trying to build good old fashioned AI style
00:35:20 models of representation of language and content of things
00:35:30 like has money.
00:35:32 So like formal logic and knowledge bases,
00:35:35 like that kind of stuff.
00:35:36 So he was integrating that with his thinking about cognition.
00:35:40 The mechanisms of cognition, how can they mechanistically
00:35:45 be applied to build these knowledge,
00:35:46 like to actually build something that
00:35:49 looks like a web of knowledge and thereby from there emerges
00:35:54 something like understanding, whatever the heck that is.
00:35:57 Yeah, he was grappling.
00:35:59 This was something that they grappled
00:36:01 with at the end of that book that I was describing,
00:36:04 Explorations in Cognition.
00:36:06 But he was realizing that the paradigm of good old fashioned
00:36:11 AI wasn’t giving him the answers to these questions.
00:36:16 By the way, that’s called good old fashioned AI now.
00:36:18 It wasn’t called that at the time.
00:36:20 Well, it was.
00:36:21 It was beginning to be called that.
00:36:23 Oh, because it was from the 60s.
00:36:24 Yeah, yeah.
00:36:26 By the late 70s, it was kind of old fashioned,
00:36:28 and it hadn’t really panned out.
00:36:30 And people were beginning to recognize that.
00:36:34 And Rommelhardt was like, yeah, he’s part of the recognition
00:36:37 that this wasn’t all working.
00:36:39 Anyway, so he started thinking in terms of the idea
00:36:48 that we needed systems that allowed us to integrate
00:36:52 multiple simultaneous constraints in a way that would
00:36:56 be mutually influencing each other.
00:37:00 So he wrote a paper that just really, first time I read it,
00:37:07 I said, oh, well, yeah, but is this important?
00:37:11 But after a while, it just got under my skin.
00:37:15 And it was called An Interactive Model of Reading.
00:37:18 And in this paper, he laid out the idea
00:37:21 that every aspect of our interpretation of what’s
00:37:34 coming off the page when we read at every level of analysis
00:37:40 you can think of actually depends
00:37:42 on all the other levels of analysis.
00:37:45 So what are the actual pixels making up each letter?
00:37:53 And what do those pixels signify about which letters they are?
00:38:00 And what do those letters tell us about what words are there?
00:38:05 And what do those words tell us about what ideas
00:38:09 the author is trying to convey?
00:38:12 And so he had this model where we
00:38:18 have these little tiny elements that represent
00:38:25 each of the pixels of each of the letters,
00:38:29 and then other ones that represent the line segments
00:38:31 in them, and other ones that represent the letters,
00:38:33 and other ones that represent the words.
00:38:36 And at that time, his idea was there’s this set of experts.
00:38:43 There’s an expert about how to construct a line out of pixels,
00:38:48 and another expert about which sets of lines
00:38:51 go together to make which letters,
00:38:53 and another one about which letters go together
00:38:55 to make which words, and another one about what
00:38:58 the meanings of the words are, and another one about how
00:39:01 the meanings fit together, and things like that.
00:39:04 And all these experts are looking at this data,
00:39:06 and they’re updating hypotheses at other levels.
00:39:12 So the word expert can tell the letter expert,
00:39:15 oh, I think there should be a T there,
00:39:17 because I think there should be a word the here.
00:39:20 And the bottom up sort of feature to letter expert
00:39:23 could say, I think there should be a T there, too.
00:39:25 And if they agree, then you see a T, right?
00:39:28 And so there’s a top down, bottom up interactive process,
00:39:32 but it’s going on at all layers simultaneously.
00:39:34 So everything can filter all the way down from the top,
00:39:37 as well as all the way up from the bottom.
00:39:39 And it’s a completely interactive, bidirectional,
00:39:42 parallel distributed process.
00:39:45 That is somehow, because of the abstractions, it’s hierarchical.
00:39:48 So there’s different layers of responsibilities,
00:39:52 different levels of responsibilities.
00:39:54 First of all, it’s fascinating to think about it
00:39:56 in this kind of mechanistic way.
00:39:58 So not thinking purely from the structure
00:40:02 of a neural network or something like a neural network,
00:40:04 but thinking about these little guys
00:40:06 that work on letters, and then the letters come words
00:40:09 and words become sentences.
00:40:11 And that’s a very interesting hypothesis
00:40:14 that from that kind of hierarchical structure
00:40:18 can emerge understanding.
00:40:21 Yeah, so, but the thing is, though,
00:40:23 I wanna just sort of relate this
00:40:25 to the earlier part of the conversation.
00:40:28 When Romelhart was first thinking about it,
00:40:31 there were these experts on the side,
00:40:34 one for the features and one for the letters
00:40:36 and one for how the letters make the words and so on.
00:40:39 And they would each be working,
00:40:43 sort of evaluating various propositions about,
00:40:46 you know, is this combination of features here
00:40:48 going to be one that looks like the letter T and so on.
00:40:52 And what he realized,
00:40:56 kind of after reading Hinton’s dissertation
00:40:59 and hearing about Jim Anderson’s
00:41:03 linear algebra based neural network models
00:41:06 that I was telling you about before
00:41:07 was that he could replace those experts
00:41:10 with neuron like processing units,
00:41:12 which just would have their connection weights
00:41:14 that would do this job.
00:41:16 So what ended up happening was
00:41:20 that Romelhart and I got together
00:41:22 and we created a model
00:41:24 called the interactive activation model of letter perception,
00:41:29 which takes these little pixel level inputs,
00:41:35 constructs line segment features, letters and words.
00:41:41 But now we built it out of a set of neuron
00:41:44 like processing units that are just connected
00:41:47 to each other with connection weights.
00:41:49 So the unit for the word time has a connection
00:41:53 to the unit for the letter T in the first position
00:41:56 and the letter I in the second position, so on.
00:41:59 And because these connections are bi directional,
00:42:05 if you have prior knowledge that it might be the word time
00:42:08 that starts to prime the letters and the features.
00:42:12 And if you don’t, then it has to start bottom up.
00:42:14 But the directionality just depends
00:42:17 on where the information comes in first.
00:42:19 And if you have context together
00:42:22 with features at the same time,
00:42:24 they can convergently result in an emergent perception.
00:42:27 And that was the piece of work that we did together
00:42:35 that sort of got us both completely convinced
00:42:41 that this neural network way of thinking
00:42:44 was going to be able to actually address the questions
00:42:48 that we were interested in as cognitive psychologists.
00:42:50 So the algorithmic side, the optimization side,
00:42:53 those are all details like when you first start the idea
00:42:56 that you can get far with this kind of way of thinking,
00:42:59 that in itself is a profound idea.
00:43:01 So do you like the term connectionism
00:43:05 to describe this kind of set of ideas?
00:43:07 I think it’s useful.
00:43:10 It highlights the notion that the knowledge
00:43:15 that the system exploits is in the connections
00:43:19 between the units, right?
00:43:21 There isn’t a separate dictionary.
00:43:24 There’s just the connections between the units.
00:43:27 So I already sort of laid that on the table
00:43:31 with the connections from the letter units
00:43:34 to the unit for the word time, right?
00:43:36 The unit for the word time isn’t a unit for the word time
00:43:40 for any other reason than it’s got the connections
00:43:43 to the letters that make up the word time.
00:43:46 Those are the units on the input that excited
00:43:48 when it’s excited that it in a sense represents
00:43:52 in the system that there’s support for the hypothesis
00:43:57 that the word time is present in the input.
00:44:01 But it’s not, the word time isn’t written anywhere
00:44:07 inside the bottle, it’s only written there
00:44:09 in the picture we drew of the model
00:44:11 to say that’s the unit for the word time, right?
00:44:14 And if somebody wants to tell me,
00:44:18 well, how do you spell that word?
00:44:21 You have to use the connections from that out
00:44:24 to then get those letters, for example.
00:44:27 That’s such a, that’s a counterintuitive idea
00:44:31 where humans want to think in this logic way.
00:44:36 This idea of connectionism, it doesn’t, it’s weird.
00:44:41 It’s weird that this is how it all works.
00:44:43 Yeah, but let’s go back to that CNN, right?
00:44:46 That CNN with all those layers of neuron
00:44:48 like processing units that we were talking about before,
00:44:51 it’s gonna come out and say, this is a cat, that’s a dog,
00:44:55 but it has no idea why it said that.
00:44:57 It’s just got all these connections
00:44:59 between all these layers of neurons,
00:45:02 like from the very first layer to the,
00:45:04 you know, like whatever these layers are,
00:45:07 they just get numbered after a while
00:45:09 because they, you know, they somehow further in you go,
00:45:13 the more abstract the features are,
00:45:17 but it’s a graded and continuous sort of process
00:45:20 of abstraction anyway.
00:45:21 And, you know, it goes from very local,
00:45:24 very specific to much more sort of global,
00:45:28 but it’s still, you know, another sort of pattern
00:45:32 of activation over an array of units.
00:45:33 And then at the output side, it says it’s a cat
00:45:36 or it’s a dog.
00:45:37 And when I open my eyes and say, oh, that’s Lex,
00:45:42 or, oh, you know, there’s my own dog
00:45:47 and I recognize my dog,
00:45:50 which is a member of the same species as many other dogs,
00:45:53 but I know this one
00:45:54 because of some slightly unique characteristics.
00:45:57 I don’t know how to describe what it is
00:46:00 that makes me know that I’m looking at Lex
00:46:02 or at my particular dog, right?
00:46:04 Or even that I’m looking at a particular brand of car.
00:46:07 Like I can say a few words about it,
00:46:09 but I wrote you a paragraph about the car,
00:46:12 you would have trouble figuring out
00:46:14 which car is he talking about, right?
00:46:16 So the idea that we have propositional knowledge
00:46:19 of what it is that allows us to recognize
00:46:23 that this is an actual instance
00:46:25 of this particular natural kind
00:46:27 has always been something that it never worked, right?
00:46:36 You couldn’t ever write down a set of propositions
00:46:38 for visual recognition.
00:46:41 And so in that space, it sort of always seemed very natural
00:46:46 that something more implicit,
00:46:51 you don’t have access to what the details
00:46:54 of the computation were in between,
00:46:56 you just get the result.
00:46:58 So that’s the other part of connectionism,
00:47:00 you cannot, you don’t read the contents of the connections,
00:47:04 the connections only cause outputs to occur
00:47:08 based on inputs.
00:47:09 Yeah, and for us that like final layer
00:47:13 or some particular layer is very important,
00:47:16 the one that tells us that it’s our dog
00:47:19 or like it’s a cat or a dog,
00:47:22 but each layer is probably equally as important
00:47:25 in the grand scheme of things.
00:47:27 Like there’s no reason why the cat versus dog
00:47:30 is more important than the lower level activations,
00:47:33 it doesn’t really matter.
00:47:34 I mean, all of it is just this beautiful stacking
00:47:36 on top of each other.
00:47:37 And we humans live in this particular layers,
00:47:40 for us it’s useful to survive,
00:47:43 to use those cat versus dog, predator versus prey,
00:47:47 all those kinds of things.
00:47:49 It’s fascinating that it’s all continuous,
00:47:51 but then you then ask,
00:47:53 the history of artificial intelligence, you ask,
00:47:55 are we able to introspect and convert the very things
00:47:59 that allow us to tell the difference between cat and dog
00:48:02 into a logic, into formal logic?
00:48:05 That’s been the dream.
00:48:06 I would say that’s still part of the dream of symbolic AI.
00:48:10 And I’ve recently talked to Doug Lenat who created Psych
00:48:19 and that’s a project that lasted for many decades
00:48:23 and still carries a sort of dream in it, right?
00:48:28 But we still don’t know the answer, right?
00:48:30 It seems like connectionism is really powerful,
00:48:34 but it also seems like there’s this building of knowledge.
00:48:38 And so how do we, how do you square those two?
00:48:41 Like, do you think the connections can contain
00:48:44 the depth of human knowledge and the depth
00:48:46 of what Dave Romahart was thinking about of understanding?
00:48:51 Well, that remains the $64 question.
00:48:55 And I…
00:48:58 With inflation, that number is higher.
00:48:59 Okay, $64,000.
00:49:01 Maybe it’s the $64 billion question now.
00:49:08 You know, I think that from the emergentist side,
00:49:13 which, you know, I placed myself on.
00:49:23 So I used to sometimes tell people
00:49:26 I was a radical, eliminative connectionist
00:49:29 because I didn’t want them to think
00:49:34 that I wanted to build like anything into the machine.
00:49:38 But I don’t like the word eliminative anymore
00:49:45 because it makes it seem like it’s wrong to think
00:49:51 that there is this emergent level of understanding.
00:49:55 And I disagree with that.
00:50:00 So I think, you know, I would call myself
00:50:02 an a radical emergentist connectionist
00:50:06 rather than eliminative connectionist, right?
00:50:09 Because I want to acknowledge
00:50:12 that these higher level kinds of aspects
00:50:17 of our cognition are real, but they’re not,
00:50:26 they don’t exist as such.
00:50:29 And there was an example that Doug Hofstadter used to use
00:50:33 that I thought was helpful in this respect.
00:50:36 Just the idea that we can think about sand dunes
00:50:42 as entities and talk about like how many there are even.
00:50:51 But we also know that a sand dune is a very fluid thing.
00:50:56 It’s a pile of sand that is capable
00:51:00 of moving around under the wind and reforming itself
00:51:08 in somewhat different ways.
00:51:10 And if we think about our thoughts as like sand dunes,
00:51:13 as being things that emerge from just the way
00:51:19 all the lower level elements sort of work together
00:51:22 and are constrained by external forces,
00:51:26 then we can say, yes, they exist as such,
00:51:29 but they also, we shouldn’t treat them
00:51:34 as completely monolithic entities that we can understand
00:51:40 without understanding sort of all of the stuff
00:51:43 that allows them to change in the ways that they do.
00:51:47 And that’s where I think the connectionist
00:51:49 feeds into the cognitive.
00:51:52 It’s like, okay, so if the substrate
00:51:55 is parallel distributed connectionist, then it doesn’t mean
00:52:01 that the contents of thought isn’t like abstract
00:52:05 and symbolic, but it’s more fluid maybe
00:52:10 than it’s easier to capture
00:52:13 with a set of logical expressions.
00:52:15 Yeah, that’s a heck of a sort of thing
00:52:17 to put at the top of a resume,
00:52:20 radical, emergentist, connectionist.
00:52:23 So there is, just like you said, a beautiful dance
00:52:26 between that, between the machinery of intelligence,
00:52:30 like the neural network side of it,
00:52:32 and the stuff that emerges.
00:52:34 I mean, the stuff that emerges seems to be,
00:52:40 I don’t know, I don’t know what that is,
00:52:44 that it seems like maybe all of reality is emergent.
00:52:48 What I think about, this is made most distinctly rich to me
00:52:57 when I look at cellular automata, look at game of life,
00:53:01 that from very, very simple things,
00:53:03 very rich, complex things emerge
00:53:06 that start looking very quickly like organisms
00:53:10 that you forget how the actual thing operates.
00:53:13 They start looking like they’re moving around,
00:53:15 they’re eating each other,
00:53:16 some of them are generating offspring.
00:53:20 You forget very quickly.
00:53:21 And it seems like maybe it’s something
00:53:23 about the human mind that wants to operate
00:53:26 in some layer of the emergent,
00:53:28 and forget about the mechanism
00:53:30 of how that emergence happens.
00:53:32 So it, just like you are in your radicalness,
00:53:35 I’m also, it seems like unfair
00:53:39 to eliminate the magic of that emergent,
00:53:43 like eliminate the fact that that emergent is real.
00:53:48 Yeah, no, I agree.
00:53:49 I’m not, that’s why I got rid of eliminative, right?
00:53:53 Eliminative, yeah.
00:53:54 Yeah, because it seemed like that was trying to say
00:53:56 that it’s all completely like.
00:54:01 An illusion of some kind, it’s not.
00:54:03 Well, who knows whether there isn’t,
00:54:06 there aren’t some illusory characteristics there.
00:54:08 And I think that philosophically many people
00:54:15 have confronted that possibility over time,
00:54:17 but it’s still important to accept it as magic, right?
00:54:26 So, I think of Fellini in this context,
00:54:30 I think of others who have appreciated the role of magic,
00:54:35 the role of magic, of actual trickery
00:54:39 in creating illusions that move us.
00:54:45 And Plato was on to this too.
00:54:47 It’s like somehow or other these shadows
00:54:52 give rise to something much deeper than that.
00:54:55 And that’s, so we won’t try to figure out what it is.
00:55:01 We’ll just accept it as given that that occurs.
00:55:04 And, you know, but he was still onto the magic of it.
00:55:08 Yeah, yeah, we won’t try to really, really,
00:55:11 really deeply understand how it works.
00:55:14 We’ll just enjoy the fact that it’s kind of fun.
00:55:16 Okay, but you worked closely with Dave Romo Hart.
00:55:21 He passed away as a human being.
00:55:24 What do you remember about him?
00:55:27 Do you miss the guy?
00:55:28 Absolutely, you know, he passed away 15ish years ago now.
00:55:38 And his demise was actually one of the most poignant
00:55:43 and, you know, like relevant tragedies, relevant to our conversation.
00:55:52 He started to undergo a progressive neurological condition
00:56:03 that isn’t far from what we’re used to.
00:56:08 A neurological condition that isn’t fully understood.
00:56:15 That is to say his particular course isn’t fully understood
00:56:23 because, you know, brain scans weren’t done at certain stages
00:56:28 and no autopsy was done or anything like that.
00:56:32 The wishes of the family.
00:56:34 We don’t know as much about the underlying pathology as we might,
00:56:38 but I had begun to get interested in this neurological condition
00:56:48 that might have been the very one that he was succumbing to
00:56:52 as my own efforts to understand another aspect of this mystery
00:56:57 that we’ve been discussing while he was beginning
00:57:01 to get progressively more and more affected.
00:57:04 So I’m going to talk about the disorder
00:57:06 and not about Rumelhart for a second, okay?
00:57:09 The disorder is something my colleagues and collaborators
00:57:12 have chosen to call semantic dementia.
00:57:17 So it’s a specific form of loss of mind
00:57:23 related to meaning, semantic dementia.
00:57:27 And it’s progressive in the sense that the patient loses the ability
00:57:37 to appreciate the meaning of the experiences that they have,
00:57:44 either from touch, from sight, from sound, from language.
00:57:50 They, I hear sounds, but I don’t know what they mean kind of thing.
00:57:56 So as this illness progresses, it starts with the patient
00:58:04 being unable to differentiate like similar breeds of dog
00:58:12 or remember the lower frequency unfamiliar categories
00:58:18 that they used to be able to remember.
00:58:21 But as it progresses, it becomes more and more striking
00:58:27 and the patient loses the ability to recognize things like
00:58:36 pigs and goats and sheep and calls all middle sized animals dogs
00:58:42 and can’t recognize rabbits and rodents anymore.
00:58:46 They call all the little ones cats
00:58:49 and they can’t recognize hippopotamuses and cows anymore.
00:58:53 They call them all horses.
00:58:55 So there was this one patient who went through this progression
00:59:00 where at a certain point, any four legged animal,
00:59:03 he would call it either a horse or a dog or a cat.
00:59:07 And if it was big, he would tend to call it a horse.
00:59:10 If it was small, he’d tend to call it a cat.
00:59:12 Middle sized ones, he called dogs.
00:59:16 This is just a part of the syndrome though.
00:59:19 The patient loses the ability to relate concepts to each other.
00:59:25 So my collaborator in this work, Carolyn Patterson,
00:59:28 developed a test called the pyramids and palm trees test.
00:59:34 So you give the patient a picture of pyramids
00:59:39 and they have a choice which goes with the pyramids,
00:59:42 palm trees or pine trees.
00:59:46 And she showed that this wasn’t just a matter of language
00:59:50 because the patient’s loss of this ability shows up
00:59:55 whether you present the material with words or with pictures.
00:59:59 The pictures, they can’t put the pictures together
01:00:03 with each other properly anymore.
01:00:05 They can’t relate the pictures to the words either.
01:00:07 They can’t do word picture matching.
01:00:09 But they’ve lost the conceptual grounding
01:00:12 from either modality of input.
01:00:15 And so that’s why it’s called semantic dementia.
01:00:19 The very semantics is disintegrating.
01:00:22 And we understand this in terms of our idea
01:00:27 that distributed representation, a pattern of activation,
01:00:31 represents the concepts, really similar ones.
01:00:33 As you degrade them, they start being,
01:00:36 you lose the differences.
01:00:40 So the difference between the dog and the goat
01:00:42 is no longer part of the pattern anymore.
01:00:44 And since dog is really familiar,
01:00:47 that’s the thing that remains.
01:00:49 And we understand that in the way the models work and learn.
01:00:52 But Rumelhart underwent this condition.
01:00:57 So on the one hand, it’s a fascinating aspect
01:01:00 of parallel distributed processing to be.
01:01:03 It reveals this sort of texture of distributed representation
01:01:08 in a very nice way, I’ve always felt.
01:01:11 But at the same time, it was extremely poignant
01:01:13 because this is exactly the condition
01:01:16 that Rumelhart was undergoing.
01:01:18 And there was a period of time when he was this man
01:01:22 who had been the most focused, goal directed, competitive,
01:01:35 thoughtful person who was willing to work for years
01:01:41 to solve a hard problem, he starts to disappear.
01:01:48 And there was a period of time when it was hard for any of us
01:01:57 to really appreciate that he was sort of, in some sense,
01:02:00 not fully there anymore.
01:02:04 Do you know if he was able to introspect
01:02:07 the solution of the understanding mind?
01:02:14 I mean, this is one of the big scientists that thinks about this.
01:02:19 Was he able to look at himself and understand the fading mind?
01:02:24 You know, we can contrast Hawking and Rumelhart in this way.
01:02:31 And I like to do that to honor Rumelhart
01:02:33 because I think Rumelhart is sort of like the Hawking
01:02:36 of cognitive science to me in some ways.
01:02:40 Both of them suffered from a degenerative condition.
01:02:45 In Hawking’s case, it affected the motor system.
01:02:49 In Rumelhart’s case, it’s affecting the semantics.
01:02:54 And not just the pure object semantics,
01:03:01 but maybe the self semantics as well.
01:03:04 And we don’t understand that.
01:03:06 Concepts broadly.
01:03:08 So I would say he didn’t.
01:03:13 And this was part of what, from the outside,
01:03:16 was a profound tragedy.
01:03:18 But on the other hand, at some level, he sort of did
01:03:22 because there was a period of time when it finally was realized
01:03:28 that he had really become profoundly impaired.
01:03:32 This was clearly a biological condition.
01:03:35 It wasn’t just like he was distracted that day or something like that.
01:03:39 So he retired from his professorship at Stanford
01:03:44 and he became, he lived with his brother for a couple years
01:03:51 and then he moved into a facility for people with cognitive impairments.
01:04:00 One that many elderly people end up in when they have cognitive impairments.
01:04:06 And I would spend time with him during that period.
01:04:12 This was like in the late 90s, around 2000 even.
01:04:16 And we would go bowling and he could still bowl.
01:04:25 And after bowling, I took him to lunch and I said,
01:04:32 where would you like to go?
01:04:34 You want to go to Wendy’s?
01:04:35 And he said, nah.
01:04:37 And I said, okay, well, where do you want to go?
01:04:38 And he just pointed.
01:04:40 He said, turn here.
01:04:41 So he still had a certain amount of spatial cognition
01:04:44 and he could get me to the restaurant.
01:04:47 And then when we got to the restaurant, I said,
01:04:51 what do you want to order?
01:04:53 And he couldn’t come up with any of the words,
01:04:56 but he knew where on the menu the thing was that he wanted.
01:04:59 So it’s, you know, and he couldn’t say what it was,
01:05:04 but he knew that that’s what he wanted to eat.
01:05:07 And so it’s like it isn’t monolithic at all.
01:05:14 Our cognition is, you know, first of all, graded in certain kinds of ways,
01:05:21 but also multipartite and there’s many elements to it and things,
01:05:27 certain sort of partial competencies still exist
01:05:31 in the absence of other aspects of these competencies.
01:05:36 So this is what always fascinated me about what used to be called
01:05:43 cognitive neuropsychology, you know,
01:05:46 the effects of brain damage on cognition.
01:05:49 But in particular, this gradual disintegration part.
01:05:53 You know, I’m a big believer that the loss of a human being that you value
01:05:59 is as powerful as, you know, first falling in love with that human being.
01:06:03 I think it’s all a celebration of the human being.
01:06:06 So the disintegration itself too is a celebration in a way.
01:06:10 Yeah, yeah.
01:06:12 But just to say something more about the scientist
01:06:17 and the backpropagation idea that you mentioned.
01:06:22 So in 1982, Hinton had been there as a postdoc and organized that conference.
01:06:34 He’d actually gone away and gotten an assistant professorship
01:06:37 and then there was this opportunity to bring him back.
01:06:41 So Jeff Hinton was back on a sabbatical.
01:06:45 San Diego.
01:06:46 And Rommelhard and I had decided we wanted to do this, you know,
01:06:52 we thought it was really exciting and the papers on the interactive activation model
01:06:58 that I was telling you about had just been published
01:07:00 and we both sort of saw a huge potential for this work and Jeff was there.
01:07:06 And so the three of us started a research group,
01:07:11 which we called the PDP Research Group.
01:07:13 And several other people came.
01:07:17 Francis Crick, who was at the Salk Institute, heard about it from Jeff
01:07:22 because Jeff was known among Brits to be brilliant
01:07:27 and Francis was well connected with his British friends.
01:07:30 So Francis Crick came.
01:07:32 That’s a heck of a group of people, wow.
01:07:34 And Paul Spolensky was one of the other postdocs.
01:07:40 He was still there as a postdoc.
01:07:41 And a few other people.
01:07:45 But anyway, Jeff talked to us about learning
01:07:56 and how we should think about how, you know, learning occurs in a neural network.
01:08:06 And he said, the problem with the way you guys have been approaching this
01:08:12 is that you’ve been looking for inspiration from biology
01:08:17 to tell you what the rules should be for how the synapses should change
01:08:22 the strengths of their connections, how the connections should form.
01:08:27 He said, that’s the wrong way to go about it.
01:08:30 What you should do is you should think in terms of
01:08:36 how you can adjust connection weights to solve a problem.
01:08:44 So you define your problem and then you figure out
01:08:49 how the adjustment of the connection weights will solve the problem.
01:08:54 And Rumelhart heard that and said to himself, okay,
01:09:01 so I’m going to start thinking about it that way.
01:09:04 I’m going to essentially imagine that I have some objective function,
01:09:11 some goal of the computation.
01:09:14 I want my machine to correctly classify all of these images.
01:09:19 And I can score that.
01:09:21 I can measure how well they’re doing on each image.
01:09:24 And I get some measure of error or loss, it’s typically called in deep learning.
01:09:30 And I’m going to figure out how to adjust the connection weights
01:09:35 so as to minimize my loss or reduce the error.
01:09:41 And that’s called, you know, gradient descent.
01:09:47 And engineers were already familiar with the concept of gradient descent.
01:09:53 And in fact, there was an algorithm called the delta rule
01:09:58 that had been invented by a professor in the electrical engineering department
01:10:07 at Stanford, Bernie Widrow and a collaborator named Hoff.
01:10:11 I never met him.
01:10:13 So gradient descent in continuous neural networks
01:10:19 with multiple neuron like processing units was already understood
01:10:26 for a single layer of connection weights.
01:10:29 We have some inputs over a set of neurons.
01:10:32 We want the output to produce a certain pattern.
01:10:35 We can define the difference between our target
01:10:38 and what the neural network is producing.
01:10:41 And we can figure out how to change the connection weights to reduce that error.
01:10:44 So what Romilhar did was to generalize that
01:10:49 so as to be able to change the connections from earlier layers of units
01:10:53 to the ones at a hidden layer between the input and the output.
01:10:58 And so he first called the algorithm the generalized delta rule
01:11:03 because it’s just an extension of the gradient descent idea.
01:11:08 And interestingly enough, Hinton was thinking that this wasn’t going to work very well.
01:11:15 So Hinton had his own alternative algorithm at the time
01:11:20 based on the concept of the Boltzmann machine that he was pursuing.
01:11:24 So the paper on the Boltzmann machine came out in,
01:11:27 learning in Boltzmann machines came out in 1985.
01:11:31 But it turned out that back prop worked better than the Boltzmann machine learning algorithm.
01:11:37 So this generalized delta algorithm ended up being called back propagation, as you say, back prop.
01:11:44 Yeah. And probably that name is opaque to me.
01:11:50 What does that mean?
01:11:53 What it meant was that in order to figure out what the changes you needed to make
01:11:59 to the connections from the input to the hidden layer,
01:12:03 you had to back propagate the error signals from the output layer
01:12:10 through the connections from the hidden layer to the output
01:12:15 to get the signals that would be the error signals for the hidden layer.
01:12:20 And that’s how Rumelhart formulated it.
01:12:22 It was like, well, we know what the error signals are at the output layer.
01:12:25 Let’s see if we can get a signal at the hidden layer
01:12:28 that tells each hidden unit what its error signal is essentially.
01:12:32 So it’s back propagating through the connections
01:12:37 from the hidden to the output to get the signals to tell the hidden units
01:12:41 how to change their weights from the input.
01:12:43 And that’s why it’s called back prop.
01:12:47 Yeah. But so it came from Hinton having introduced the concept of, you know,
01:12:54 define your objective function, figure out how to take the derivative
01:12:59 so that you can adjust the connections so that they make progress towards your goal.
01:13:04 So stop thinking about biology for a second
01:13:06 and let’s start to think about optimization and computation a little bit more.
01:13:12 So what about Jeff Hinton?
01:13:15 You’ve gotten a chance to work with him in that little thing.
01:13:20 The set of people involved there is quite incredible.
01:13:24 The small set of people under the PDP flag,
01:13:28 it’s just given the amount of impact those ideas have had over the years,
01:13:32 it’s kind of incredible to think about.
01:13:34 But, you know, just like you said, like yourself,
01:13:38 Jeffrey Hinton is seen as one of the, not just like a seminal figure in AI,
01:13:43 but just a brilliant person,
01:13:45 just like the horsepower of the mind is pretty high up there for him
01:13:49 because he’s just a great thinker.
01:13:52 So what kind of ideas have you learned from him?
01:13:57 Have you influenced each other on?
01:13:59 Have you debated over what stands out to you in the full space of ideas here
01:14:05 at the intersection of computation and cognition?
01:14:09 Well, so Jeff has said many things to me that had a profound impact on my thinking.
01:14:18 And he’s written several articles which were way ahead of their time.
01:14:26 He had two papers in 1981, just to give one example,
01:14:37 one of which was essentially the idea of transformers
01:14:42 and another of which was an early paper on semantic cognition
01:14:49 which inspired him and Rumelhart and me throughout the 80s
01:15:01 and, you know, still I think sort of grounds my own thinking
01:15:11 about the semantic aspects of cognition.
01:15:16 He also, in a small paper that was never published that he wrote in 1977,
01:15:25 you know, before he actually arrived at UCSD or maybe a couple years even before that,
01:15:29 I don’t know, when he was a PhD student,
01:15:32 he described how a neural network could do recursive computation.
01:15:40 And it was a very clever idea that he’s continued to explore over time,
01:15:48 which was sort of the idea that when you call a subroutine,
01:15:56 you need to save the state that you had when you called it
01:16:01 so you can get back to where you were when you’re finished with the subroutine.
01:16:04 And the idea was that you would save the state of the calling routine
01:16:10 by making fast changes to connection weights.
01:16:13 And then when you finished with the subroutine call,
01:16:19 those fast changes in the connection weights would allow you to go back
01:16:23 to where you had been before and reinstate the previous context
01:16:27 so that you could continue on with the top level of the computation.
01:16:32 Anyway, that was part of the idea.
01:16:35 And I always thought, okay, that’s really, you know,
01:16:38 he had extremely creative ideas that were quite a lot ahead of his time
01:16:44 and many of them in the 1970s and early 1980s.
01:16:49 So another thing about Geoff Hinton’s way of thinking,
01:16:57 which has profoundly influenced my effort to understand
01:17:05 human mathematical cognition, is that he doesn’t write too many equations.
01:17:13 And people tell stories like, oh, in the Hinton Lab meetings,
01:17:17 you don’t get up at the board and write equations
01:17:19 like you do in everybody else’s machine learning lab.
01:17:22 What you do is you draw a picture.
01:17:26 And, you know, he explains aspects of the way deep learning works
01:17:33 by putting his hands together and showing you the shape of a ravine
01:17:38 and using that as a geometrical metaphor for what’s happening
01:17:45 as this gradient descent process.
01:17:47 You’re coming down the wall of a ravine.
01:17:49 If you take too big a jump, you’re going to jump to the other side.
01:17:53 And so that’s why we have to turn down the learning rate, for example.
01:17:59 And it speaks to me of the fundamentally intuitive character of deep insight
01:18:12 together with a commitment to really understanding
01:18:21 in a way that’s absolutely ultimately explicit and clear, but also intuitive.
01:18:31 Yeah, there’s certain people like that.
01:18:33 Here’s an example, some kind of weird mix of visual and intuitive
01:18:38 and all those kinds of things.
01:18:40 Feynman is another example, different style of thinking, but very unique.
01:18:44 And when you’re around those people, for me in the engineering realm,
01:18:48 there’s a guy named Jim Keller who’s a chip designer, engineer.
01:18:52 Every time I talk to him, it doesn’t matter what we’re talking about.
01:18:57 Just having experienced that unique way of thinking transforms you
01:19:02 and makes your work much better.
01:19:04 And that’s the magic.
01:19:06 You look at Daniel Kahneman, you look at the great collaborations
01:19:10 throughout the history of science.
01:19:12 That’s the magic of that.
01:19:13 It’s not always the exact ideas that you talk about,
01:19:16 but it’s the process of generating those ideas.
01:19:19 Being around that, spending time with that human being,
01:19:22 you can come up with some brilliant work,
01:19:24 especially when it’s cross disciplinary as it was a little bit in your case with Jeff.
01:19:29 Yeah.
01:19:31 Jeff is a descendant of the logician Boole.
01:19:38 He comes from a long line of English academics.
01:19:43 And together with the deeply intuitive thinking ability that he has,
01:19:51 he also has, it’s been clear, he’s described this to me,
01:19:59 and I think he’s mentioned it from time to time in other interviews
01:20:04 that he’s had with people.
01:20:06 He’s wanted to be able to sort of think of himself as contributing
01:20:12 to the understanding of reasoning itself, not just human reasoning.
01:20:22 Like Boole is about logic, right?
01:20:25 It’s about what can we conclude from what else and how do we formalize that.
01:20:31 And as a computer scientist, logician, philosopher,
01:20:40 the goal is to understand how we derive truths from other,
01:20:46 from givens and things like this.
01:20:48 And the work that Jeff was doing in the early to mid 80s
01:20:57 on something called the Bolton machine was his way of connecting
01:21:02 with that Boolean tradition and bringing it into the more continuous,
01:21:07 probabilistic graded constraint satisfaction realm.
01:21:11 And it was a beautiful set of ideas linked with theoretical physics
01:21:20 as well as with logic.
01:21:26 And it’s always been, I mean, I’ve always been inspired
01:21:31 by the Bolton machine too.
01:21:33 It’s like, well, if the neurons are probabilistic rather than deterministic
01:21:38 in their computations, then maybe this somehow is part of the serendipity
01:21:48 or adventitiousness of the moment of insight, right?
01:21:53 It might not have occurred at that particular instant.
01:21:56 It might be sort of partially the result of a stochastic process.
01:22:00 And that too is part of the magic of the emergence of some of these things.
01:22:07 Well, you’re right with the Boolean lineage and the dream of computer science
01:22:11 is somehow, I mean, I certainly think of humans this way,
01:22:16 that humans are one particular manifestation of intelligence,
01:22:20 that there’s something bigger going on and you’re hoping to figure that out.
01:22:25 The mechanisms of intelligence, the mechanisms of cognition
01:22:28 are much bigger than just humans.
01:22:30 Yeah. So I think of, I started using the phrase computational intelligence
01:22:37 at some point as to characterize the field that I thought, you know,
01:22:42 people like Geoff Hinton and many of the people I know at DeepMind
01:22:51 are working in and where I feel like I’m, you know,
01:23:00 I’m a kind of a human oriented computational intelligence researcher
01:23:06 in that I’m actually kind of interested in the human solution.
01:23:10 But at the same time, I feel like that’s where a huge amount
01:23:18 of the excitement of deep learning actually lies is in the idea that,
01:23:26 you know, we may be able to even go beyond what we can achieve
01:23:32 with our own nervous systems when we build computational intelligences
01:23:38 that are, you know, not limited in the ways that we are by our own biology.
01:23:46 Perhaps allowing us to scale the very mechanisms of human intelligence
01:23:51 just increases power through scale.
01:23:55 Yes. And I think that that, you know, obviously that’s the,
01:24:03 that’s being played out massively at Google Brain, at OpenAI
01:24:08 and to some extent at DeepMind as well.
01:24:11 I guess I shouldn’t say to some extent.
01:24:14 Just the massive scale of the computations that are used to succeed
01:24:22 at games like Go or to solve the protein folding problems
01:24:25 that they’ve been solving and so on.
01:24:27 Still not as many synapses and neurons as the human brain.
01:24:31 So we still got, we’re still beating them on that.
01:24:35 We humans are beating the AIs, but they’re catching up pretty quickly.
01:24:41 You write about modeling of mathematical cognition.
01:24:45 So let me first ask about mathematics in general.
01:24:49 There’s a paper titled Parallel Distributed Processing
01:24:53 Approach to Mathematical Cognition where in the introduction
01:24:56 there’s some beautiful discussion of mathematics.
01:25:00 And you referenced there Tristan Needham who criticizes a narrow
01:25:05 form of view of mathematics by liking the studying of mathematics
01:25:10 as symbol manipulation to studying music without ever hearing a note.
01:25:16 So from that perspective, what do you think is mathematics?
01:25:20 What is this world of mathematics like?
01:25:23 Well, I think of mathematics as a set of tools for exploring
01:25:32 idealized worlds that often turn out to be extremely relevant
01:25:42 to the real world but need not.
01:25:47 But they’re worlds in which objects exist with idealized properties
01:26:01 and in which the relationships among them can be characterized
01:26:07 with precision so as to allow the implications of certain facts
01:26:17 to then allow you to derive other facts with certainty.
01:26:22 So if you have two triangles and you know that there is an angle
01:26:37 in the first one that has the same measure as an angle in the second one
01:26:42 and you know that the lengths of the sides adjacent to that angle
01:26:47 in each of the two triangles, the corresponding sides adjacent
01:26:53 to that angle also have the same measure, then you can then conclude
01:26:58 that the triangles are congruent.
01:27:02 That is to say they have all of their properties in common.
01:27:06 And that is something about triangles.
01:27:11 It’s not a matter of formulas.
01:27:15 These are idealized objects.
01:27:18 In fact, we built bridges out of triangles and we understand
01:27:26 how to measure the height of something we can’t climb by extending
01:27:32 these ideas about triangles a little further.
01:27:36 And all of the ability to get a tiny speck of matter launched
01:27:49 from the planet Earth to intersect with some tiny, tiny little body
01:27:56 way out in way beyond Pluto somewhere at exactly a predicted time
01:28:02 and date is something that depends on these ideas.
01:28:08 And it’s actually happening in the real physical world that these ideas
01:28:18 make contact with it in those kinds of instances.
01:28:27 But there are these idealized objects, these triangles or these distances
01:28:32 or these points, whatever they are, that allow for this set of tools
01:28:40 to be created that then gives human beings this incredible leverage
01:28:47 that they didn’t have without these concepts.
01:28:51 And I think this is actually already true when we think about just,
01:29:01 you know, the natural numbers.
01:29:06 I always like to include zero, so I’m going to say the nonnegative integers,
01:29:11 but that’s a place where some people prefer not to include zero.
01:29:17 We like zero here, natural numbers, zero, one, two, three, four, five,
01:29:21 six, seven, and so on.
01:29:23 Yeah. And because they give you the ability to be exact about
01:29:36 how many sheep you have.
01:29:38 I sent you out this morning, there were 23 sheep.
01:29:41 You came back with only 22. What happened?
01:29:44 The fundamental problem of physics, how many sheep you have.
01:29:48 It’s a fundamental problem of human society that you damn well better
01:29:53 bring back the same number of sheep as you started with.
01:29:57 And it allows commerce, it allows contracts, it allows the establishment
01:30:03 of records and so on to have systems that allow these things to be notated.
01:30:10 But they have an inherent aboutness to them that’s one in the same time sort of
01:30:20 abstract and idealized and generalizable, while on the other hand,
01:30:26 potentially very, very grounded and concrete.
01:30:30 And one of the things that makes for the incredible achievements of the human mind
01:30:41 is the fact that humans invented these idealized systems that leverage
01:30:49 the power of human thought in such a way as to allow all this kind of thing to happen.
01:30:57 And so that’s what mathematics to me is the development of systems for thinking about
01:31:06 the properties and relations among sets of idealized objects and
01:31:18 the mathematical notation system that we unfortunately focus way too much on
01:31:26 is just our way of expressing propositions about these properties.
01:31:36 It’s just like we’re talking with Chomsky in language.
01:31:39 It’s the thing we’ve invented for the communication of those ideas.
01:31:43 They’re not necessarily the deep representation of those ideas.
01:31:48 So what’s a good way to model such powerful mathematical reasoning, would you say?
01:31:57 What are some ideas you have for capturing this in a model?
01:32:01 The insights that human mathematicians have had is a combination of the kind of the
01:32:10 intuitive kind of connectionist like knowledge that makes it so that something is just like
01:32:24 obviously true so that you don’t have to think about why it’s true.
01:32:31 That then makes it possible to then take the next step and ponder and reason and
01:32:40 figure out something that you previously didn’t have that intuition about.
01:32:45 It then ultimately becomes a part of the intuition that the next generation of
01:32:54 mathematical thinkers have to ground their own thinking on so that they can extend the ideas even further.
01:33:02 I came across this quotation from Henri Poincare while I was walking in the woods with my wife
01:33:15 in a state park in Northern California late last summer.
01:33:20 And what it said on the bench was it is by logic that we prove but by intuition that we discover.
01:33:32 And so what for me the essence of the project is to understand how to bring the intuitive
01:33:41 connectionist resources to bear on letting the intuitive discovery arise from engagement in
01:33:56 thinking with this formal system.
01:33:59 So I think of the ability of somebody like Hinton or Newton or Einstein or Rumelhart or
01:34:14 Poincare to Archimedes is another example.
01:34:21 So suddenly a flash of insight occurs. It’s like the constellation of all of these
01:34:31 simultaneous constraints that somehow or other causes the mind to settle into a novel state that
01:34:38 it never did before and give rise to a new idea that then you can say, okay, well, now how can I
01:34:51 prove this? How do I write down the steps of that theorem that allow me to make it rigorous and certain?
01:35:01 And so I feel like the kinds of things that we’re beginning to see deep learning systems do of
01:35:14 their own accord kind of gives me this feeling of hope or encouragement that ultimately it’ll all happen.
01:35:34 So in particular as many people now have become really interested in thinking about, you know,
01:35:46 neural networks that have been trained with massive amounts of text can be given a prompt and they
01:35:55 can then sort of generate some really interesting, fanciful, creative story from that prompt.
01:36:05 And there’s kind of like a sense that they’ve somehow synthesized something like novel out of
01:36:15 the, you know, all of the particulars of all of the billions and billions of experiences that went
01:36:22 into the training data that gives rise to something like this sort of intuitive sense of what would
01:36:29 be a fun and interesting little story to tell or something like that. It just sort of wells up out
01:36:36 of the letting the thing play out its own imagining of what somebody might say given this prompt as
01:36:47 an input to get it to start to generate its own thoughts. And to me that sort of represents the
01:36:56 potential of capturing the intuitive side of this.
01:37:01 And there’s other examples, I don’t know if you find them as captivating is, you know, on the
01:37:06 DeepMind side with AlphaZero, if you study chess, the kind of solutions that has come up in terms
01:37:12 of chess, it is, there’s novel ideas there. It feels very like there’s brilliant moments of insight.
01:37:20 And the mechanism they use, if you think of search as maybe more towards good old fashioned AI and
01:37:31 then there’s the connection is the neural network that has the intuition of looking at a board,
01:37:37 looking at a set of patterns and saying, how good is this set of positions? And the next few
01:37:42 positions, how good are those? And that’s it. That’s just an intuition. Grandmasters have this
01:37:49 and understanding positionally, tactically, how good the situation is, how can it be improved
01:37:55 without doing this full, like deep search. And then maybe doing a little bit of what human chess
01:38:03 players call calculation, which is the search, taking a particular set of steps down the line to
01:38:08 see how they unroll. But there is moments of genius in those systems too. So that’s another hopeful
01:38:16 illustration that from neural networks can emerge this novel creation of an idea.
01:38:25 Yes. And I think that, you know, I think Demis Hassabis is, you know, he’s spoken about those
01:38:34 things. I heard him describe a move that was made in one of the go matches against Lisa Dahl in a
01:38:44 very similar way. And it caused me to become really excited to kind of collaborate with some of those
01:38:52 people and analyze it at DeepMind. So I think though that what I like to really emphasize here
01:39:05 is one part of what I like to emphasize about mathematical cognition at least is that philosophers
01:39:15 and logicians going back three or even a little more than 3000 years ago began to develop these
01:39:28 formal systems and gradually the whole idea about thinking formally got constructed. And, you know,
01:39:45 it’s preceded Euclid, certainly present in the work of Thales and others. And I’m not the world’s
01:39:55 leading expert in all the details of that history, but Euclid’s elements were the kind of the touch
01:40:03 point of a coherent document that sort of laid out this idea of an actual formal system within which
01:40:15 these objects were characterized and the system of inference that allowed new truths to be derived
01:40:31 from others was sort of like established as a paradigm. And what I find interesting is the
01:40:43 idea that the ability to become a person who is capable of thinking in this abstract formal way
01:40:55 is a result of the same kind of immersion in experience thinking in that way that we now
01:41:10 begin to think of our understanding of language as being, right? So, we immerse ourselves in a
01:41:16 particular language, in a particular world of objects and their relationships and we learn
01:41:22 to talk about that and we develop intuitive understanding of the real world. In a similar
01:41:30 way, we can think that what academia has created for us, what those early philosophers and their
01:41:39 academies in Athens and Alexandria and other places allowed was the development of these
01:41:49 schools of thought, modes of thought that then become deeply ingrained and it becomes what it
01:42:00 is that makes it so that somebody like Jerry Fodor would think that systematic thought is
01:42:11 the essential characteristic of the human mind as opposed to a derived and an acquired characteristic
01:42:20 that results from acculturation in a certain mode that’s been invented by humans.
01:42:28 Would you say it’s more fundamental than like language? If we start dancing, if we bring
01:42:34 Chomsky back into the conversation, first of all, is it unfair to draw a line between mathematical
01:42:43 cognition and language, linguistic cognition?
01:42:48 I think that’s a very interesting question and I think it’s one of the ones that I’m actually very
01:42:54 interested in right now, but I think the answer is in important ways, it is important to draw that
01:43:06 line, but then to come back and look at it again and see some of the subtleties and interesting
01:43:12 aspects of the difference. So if we think about Chomsky himself, he was born into an academic
01:43:34 family. His father was a professor of rabbinical studies at a small rabbinical college in
01:43:40 Philadelphia. He was deeply enculturated in a culture of thought and reason and brought to the
01:43:59 effort to understand natural language, this profound engagement with these formal systems. I
01:44:13 think that there was tremendous power in that and that Chomsky had some amazing insights into the
01:44:23 structure of natural language, but that, I’m going to use the word but there, the actual intuitive
01:44:34 knowledge of these things only goes so far and does not go as far as it does in people like
01:44:41 Chomsky himself. And this was something that was discovered in the PhD dissertation of Lyla
01:44:48 Gleitman, who was actually trained in the same linguistics department with Chomsky. So what Lyla
01:44:55 discovered was that the intuitions that linguists had about even the meaning of a phrase, not just
01:45:09 about its grammar, but about what they thought a phrase must mean were very different from the
01:45:17 intuitions of an ordinary person who wasn’t a formally trained thinker. And well, it recently
01:45:27 has become much more salient. I happened to have learned about this when I myself was a PhD student
01:45:32 at the University of Pennsylvania, but I never knew how to put it together with all of my other
01:45:38 thinking about these things. So I actually currently have the hypothesis that formally
01:45:45 trained linguists and other formally trained academics, whether it be linguistics, philosophy,
01:45:58 cognitive science, computer science, machine learning, mathematics,
01:46:02 have a mode of engagement with experience that is intuitively deeply structured to be more
01:46:17 organized around the systematicity and ability to be conformant with the principles of a system
01:46:35 than is actually true of the natural human mind without that immersion.
01:46:42 That’s fascinating. So the different fields and approaches with which you start to study the mind
01:46:48 actually take you away from the natural operation of the mind. So it makes it very difficult for you
01:46:56 to be somebody who introspects.
01:46:59 Yes. And this is where things about human belief and so called knowledge that we consider
01:47:16 private, not our business to manipulate in others. We are not entitled to tell somebody else what to
01:47:29 believe about certain kinds of things. What are those beliefs? Well, they are the product of this
01:47:42 sort of immersion and enculturation. That is what I believe.
01:47:51 And that’s limiting.
01:47:55 It’s something to be aware of.
01:47:58 Does that limit you from having a good model of cognition?
01:48:04 It can.
01:48:04 So when you look at mathematical or linguistics, I mean, what is that line then? So is Chomsky
01:48:13 unable to sneak up to the full picture of cognition? Are you, when you’re focusing on
01:48:17 mathematical thinking, are you also unable to do so?
01:48:22 I think you’re right. I think that’s a great way of characterizing it. And
01:48:27 I also think that it’s related to the concept of beginner’s mind and another concept called the
01:48:43 expert blind spot. So the expert blind spot is much more prosaic seeming than this point that
01:48:53 you were just making. But it’s something that plagues experts when they try to communicate
01:49:01 their understanding to non experts. And that is that things are self evident to them that
01:49:12 they can’t begin to even think about how they could explain it to somebody else.
01:49:23 Because it’s just like so patently obvious that it must be true. And
01:49:31 when Kronacker said, God made the natural numbers, all else is the work of man,
01:49:47 he was expressing that intuition that somehow or other, the basic fundamentals of discrete
01:49:57 quantities being countable and innumerable and indefinite in number was not something that
01:50:10 had to be discovered. But he was wrong. It turns out that many cognitive scientists
01:50:21 agreed with him for a time. There was a long period of time where the natural
01:50:27 numbers were considered to be a part of the innate endowment of core knowledge or to use
01:50:35 the kind of phrases that Spelke and Kerry used to talk about what they believe are
01:50:41 the innate primitives of the human mind. And they no longer believe that. It’s actually
01:50:50 been more or less accepted by almost everyone that the natural numbers are actually a cultural
01:50:56 construction. And it’s so interesting to go back and study those few people who still exist who
01:51:04 don’t have those systems. So this is just an example to me where a certain mode of thinking
01:51:13 about language itself or a certain mode of thinking about geometry and those kinds of
01:51:20 relations. So it becomes so second nature that you don’t know what it is that you need to teach. And
01:51:30 in fact, we don’t really teach it all that explicitly anyway. You take a math class,
01:51:41 the professor sort of teaches it to you the way they understand it. Some of the students in the
01:51:47 class sort of like they get it. They start to get the way of thinking and they can actually do the
01:51:52 problems that get put on the homework that the professor thinks are interesting and challenging
01:51:57 ones. But most of the students who don’t kind of engage as deeply don’t ever get. And we think,
01:52:08 oh, that man must be brilliant. He must have this special insight. But he must have some
01:52:14 some biological sort of bit that’s different, that makes him so that he or she could have
01:52:20 that insight. Although I don’t want to dismiss biological individual differences completely,
01:52:31 I find it much more interesting to think about the possibility that it was that difference in the
01:52:39 dinner table conversation at the Chomsky house when he was growing up that made it so that he
01:52:45 had that cast of mind. Yeah. And there’s a few topics we talked about that kind of interconnect
01:52:53 because I wonder the better I get at certain things, we humans, the deeper we understand
01:52:59 something, what are you starting to then miss about the rest of the world? We talked about David
01:53:11 and his degenerative mind. And, you know, when you look in the mirror and wonder how different
01:53:19 am I am I cognitively from the man I was a month ago, from the man I was a year ago, like what,
01:53:26 you know, if I can, having thought about language of Chomsky for 10, 20 years, what am I no longer
01:53:35 able to see? What is in my blind spot? And how big is that? And then to somehow be able to leap back
01:53:43 out of your deep, like structure that you form for yourself about thinking about the world,
01:53:48 leap back and look at the big picture again, or jump out of the your current way of thinking.
01:53:54 And to be able to introspect, like what are the limitations of your mind? How is your mind less
01:54:00 powerful than it used to be or more powerful or different, powerful in different ways? So that
01:54:06 seems to be a difficult thing to do because we’re living, we’re looking at the world through the
01:54:11 lens of our mind, right? To step outside and introspect is difficult, but it seems necessary
01:54:17 if you want to make progress. You know, one of the threads of psychological research that’s always
01:54:25 been very, I don’t know, important to me to be aware of is the idea that our explanations of our
01:54:38 own behavior aren’t necessarily actually part of the causal process that caused that behavior to
01:54:53 occur, or even valid observations of the set of constraints that led to the outcome, but they are
01:55:03 post hoc rationalizations that we can give based on information at our disposal about what might
01:55:11 have contributed to the result that we came to when asked. And so this is an idea that was
01:55:21 introduced in a very important paper by Nisbet and Wilson about, you know, the limits on our ability
01:55:29 to be aware of the factors that cause us to make the choices that we make. And, you know, I think
01:55:42 it’s something that we really ought to be much more cognizant of, in general, as human beings,
01:55:54 is that our own insight into exactly why we hold the beliefs that we do and we hold the attitudes
01:56:01 and make the choices and feel the feelings that we do is not something that we totally control
01:56:12 or totally observe. And it’s subject to, you know, our culturally transmitted understanding of what
01:56:25 it is that is the mode that we give to explain these things when asked to do so as much as it is
01:56:34 about anything else. And so even our ability to introspect and think we have access to our own
01:56:42 thoughts is a product of culture and belief, you know, practice.
01:56:47 So let me ask you the big question of advice. So you’ve lived an incredible life in terms of the
01:56:57 ideas you’ve put out into the world, in terms of the trajectory you’ve taken through your career,
01:57:02 through your life. What advice would you give to young people today, in high school, in college,
01:57:09 about how to have a career or how to have a life they can be proud of?
01:57:16 Finding the thing that you are intrinsically motivated to engage with and then celebrating
01:57:27 that discovery is what it’s all about. When I was in college, I struggled with that. I had thought
01:57:43 I wanted to be a psychiatrist because I think I was interested in human psychology in high school.
01:57:50 And at that time, the only sort of information I had that had anything to do with the psyche was,
01:57:58 you know, Freud and Erich Fromm and sort of popular psychiatry kinds of things.
01:58:03 And so, well, they were psychiatrists, right? So I had to be a psychiatrist.
01:58:08 And that meant I had to go to medical school. And I got to college and I find myself taking,
01:58:14 you know, the first semester of a three quarter physics class and it was mechanics. And this was
01:58:21 so far from what it was I was interested in, but it was also too early in the morning in the winter
01:58:26 court semester. So I never made it to the physics class. But I wondered about the rest of my
01:58:34 freshman year and most of my sophomore year until I found myself in the midst of this situation where
01:58:45 around me there was this big revolution happening. I was at Columbia University in 1968 and
01:58:54 the Vietnam War is going on. Columbia is building a gym in Morningside Heights, which is part of
01:58:59 Harlem. And people are thinking, oh, the big bad rich guys are stealing the parkland that
01:59:06 belongs to the people of Harlem. And, you know, they’re part of the military industrial complex,
01:59:13 which is enslaving us and sending us all off to war in Vietnam. And so there was a big revolution
01:59:20 that involved a confluence of black activism and, you know, SDS and social justice and the whole
01:59:27 university blew up and got shut down. And I got a chance to sort of think about
01:59:34 why people were behaving the way they were in this context. And I, you know, I happened to have
01:59:42 taken mathematical statistics. I happened to have been taking psychology that quarter at just cycle
01:59:48 one. And somehow things in that space all ran together in my mind and got me really excited
01:59:54 about asking questions about why people, what made certain people go into the buildings and not
02:00:01 others and things like that. And so suddenly I had a path forward and I had just been wandering
02:00:07 around aimlessly. And at the different points in my career, you know, and I think, okay,
02:00:12 well, should I take this class or should I just read that book about some idea that I want to
02:00:26 understand better, you know, or should I pursue the thing that excites me and interests me or
02:00:33 should I, you know, meet some requirement? You know, that’s, I always did the latter.
02:00:39 So I ended up, my professors in psychology were, thought I was great. They wanted me to go to
02:00:46 graduate school. They nominated me for Phi Beta Kappa. And I went to the Phi Beta Kappa ceremony
02:00:55 and this guy came up and he said, oh, are you Magna Arsuma? And I wasn’t even getting honors
02:01:00 based on my grades. They just happened to have thought I was interested enough in ideas to
02:01:07 belong to Phi Beta Kappa. So. I mean, would it be fair to say you kind of stumbled around a little
02:01:12 bit through accidents of too early morning of classes in physics and so on until you discovered
02:01:20 intrinsic motivation, as you mentioned, and then that’s it. It hooked you. And then you celebrate
02:01:26 the fact that this happens to human beings. Yeah. And what is it that made what I did intrinsically
02:01:34 motivating to me? Well, that’s interesting and I don’t know all the answers to it. And I don’t
02:01:41 think I want anybody to think that you should be sort of in any way, I don’t know, sanctimonious or
02:01:52 anything about it. You know, it’s like, I really enjoyed doing statistical analysis of data. I
02:02:01 really enjoyed running my own experiment, which was what I got a chance to do in the psychology
02:02:09 department that chemistry and physics had never, I never imagined that mere mortals would ever do
02:02:14 an experiment in those sciences, except one that was in the textbook that you were told to do in
02:02:20 lab class. But in psychology, we were already like, even when I was taking psych one, it turned out
02:02:26 we had our own rat and we got to, after two set experiments, we got to, okay, do something you
02:02:32 think of with your rat. So it’s the opportunity to do it myself and to bring together a certain
02:02:42 set of things that engaged me intrinsically. And I think it has something to do with why
02:02:49 certain people turn out to be profoundly amazing musical geniuses, right? They get immersed in it
02:02:59 at an early enough point and it just sort of gets into the fabric. So my little brother had intrinsic
02:03:07 motivation for music as we witnessed when he discovered how to put records on the phonograph
02:03:15 when he was like 13 months old and recognize which one he wanted to play, not because he could read
02:03:21 the labels, because he could sort of see which ones had which scratches, which were the different,
02:03:26 you know, oh, that’s rapidi espanol. And that’s, you know, and, and, and,
02:03:31 And he enjoyed that, that connected with him somehow.
02:03:33 Yeah. And, and there was something that it fed into and it, you’re extremely lucky if you have
02:03:40 that and if you can nurture it and can let it grow and let it be, be an important part of your life.
02:03:47 Yeah. Those are, those are the two things is like, be attentive enough to,
02:03:52 to feel it when it comes, like this is something special. I mean, I don’t know. For example,
02:03:59 I really like tabular data, like Excel sheets. Like it brings me a deep joy. I don’t know how
02:04:08 useful that is for anything. That’s part of what I’m talking about.
02:04:12 Exactly. So there’s like a million, not a million, but there’s a lot of things
02:04:17 like that. For me, you have to hear that for yourself, like be, like realize this is really
02:04:23 joyful. But then the other part that you’re mentioning, which is the nurture is take time
02:04:27 and stay with it, stay with it a while and see where that takes you in life.
02:04:33 Yeah. And I think, I think the, the, the motivational engagement results in the
02:04:40 immersion that then creates the opportunity to obtain the expertise. So, you know, we could call
02:04:47 it the Mozart effect, right? I mean, when I think about Mozart, I think about, you know,
02:04:53 the person who was born as the fourth member of the family string quartet, right? And, and they
02:05:01 handed him the violin when he was six weeks old. All right, start playing, you know, it’s like,
02:05:08 and so the, the level of immersion there was, was amazingly profound, but hopefully he also had,
02:05:20 you know, some, something, maybe this is where the more sort of the genetic part comes in.
02:05:28 Sometimes I think, you know, something in him resonated to the music so that that,
02:05:34 the synergy of the combination of that was so powerful. So, so that’s what I really considered
02:05:40 to be the Mozart effect. It’s sort of the, the synergy of something with, with experience that,
02:05:47 that then results in the unique flowering of a particular, you know, mind.
02:05:51 And so I know my siblings and I are all very different from each other. We’ve all gone in
02:06:01 our own different directions. And, you know, I mentioned my younger brother who was very musical.
02:06:07 I had my other younger brother was like this amazing, like intuitive engineer.
02:06:11 And my sister, one of my sisters was passionate about, in, you know, water conservation well
02:06:23 before it was, you know, such a hugely important issue that it is today. So we all sort of somehow
02:06:31 these find a different thing. And I don’t, I don’t mean to say it isn’t tied in with something about,
02:06:41 about us biologically, but, but it’s also when that happens, where you can find that, then,
02:06:47 you know, you can do your thing and you can be excited about it. So people can be excited about
02:06:52 fitting people on bicycles, as well as excited about making neural networks, achieve insights
02:06:56 into human cognition, right? Yeah. Like for me personally, I’ve always been excited about
02:07:03 love and friendship between humans. And just like the actual experience of it,
02:07:10 since I was a child, just observing people around me and also been excited about robots.
02:07:16 And there’s something in me that thinks I really would love to explore how those two things
02:07:21 combine. And it doesn’t make any sense. A lot of it is also timing, just to think of your own career
02:07:26 and your own life. You found yourself in certain pieces, places that happened to involve some of
02:07:33 the greatest thinkers of our time. And so it just worked out that like, you guys developed those
02:07:37 ideas. And there may be a lot of other people similar to you, and they were brilliant, and
02:07:43 they never found that right connection and place to where they, their ideas could flourish. So
02:07:48 it’s timing, it’s place, it’s people. And ultimately the whole ride, you know, it’s undirected.
02:07:56 Can I ask you about something you mentioned in terms of psychiatry when you were younger?
02:08:00 Because I had a similar experience of, you know, reading Freud and Carl Jung and just,
02:08:09 you know, those kind of popular psychiatry ideas. And that was a dream for me early on in high
02:08:15 school too. Like I hoped to understand the human mind by, somehow psychiatry felt like
02:08:24 the right discipline for that. Does that make you sad? That psychiatry is not
02:08:31 the mechanism by which you are able to explore the human mind. So for me, I was a little bit
02:08:37 disillusioned because of how much prescription medication and biochemistry is involved in the
02:08:46 discipline of psychiatry, as opposed to the dream of the Freud like, use the mechanisms of language
02:08:53 to explore the human mind. So that was a little disappointing. And that’s why I kind of went to
02:09:00 computer science and thinking like, maybe you can explore the human mind by trying to build the
02:09:04 thing. Yes. I wasn’t exposed to the sort of the biomedical slash pharmacological aspects of
02:09:14 psychiatry at that point because I dropped out of that whole idea of premed that I never even
02:09:22 found out about that until much later. But you’re absolutely right. So I was actually a member of the
02:09:30 National Advisory Mental Health Council. That is to say the board of scientists who advise the
02:09:41 director of the National Institute of Mental Health. And that was around the year 2000. And
02:09:47 in fact, at that time, the man who came in as the new director, I had been on this board for a year
02:09:56 when he came in, said, okay, schizophrenia is a biological illness. It’s a lot like cancer.
02:10:08 We’ve made huge strides in curing cancer. And that’s what we’re going to do with schizophrenia.
02:10:13 We’re going to find the medications that are going to cure this disease. And we’re not going
02:10:18 to listen to anybody’s grandmother anymore. And good old behavioral psychology is not something
02:10:27 we’re going to support any further. And he completely alienated me from the Institute
02:10:40 and from all of its prior policies, which had been much more holistic, I think, really at some level.
02:10:46 And the other people on the board were like psychiatrists, very biological psychiatrists.
02:10:57 It didn’t pan out that nothing has changed in our ability to help people with mental illness.
02:11:07 And so 20 years later, that particular path was a dead end, as far as I can tell.
02:11:14 Well, there’s some aspect to, and sorry to romanticize the whole philosophical conversation
02:11:20 about the human mind. But to me, psychiatrists, for a time, held the flag of we’re the deep thinkers.
02:11:29 In the same way that physicists are the deep thinkers about the nature of reality,
02:11:34 psychiatrists are the deep thinkers about the nature of the human mind. And I think that flag
02:11:38 has been taken from them and carried by people like you. It’s like, it’s more in the cognitive
02:11:44 psychology, especially when you have a foot in the computational view of the world, because you can
02:11:50 both build it, you can like, intuit about the functioning of the mind by building little models
02:11:56 and be able to see mathematical things and then deploying those models, especially in computers,
02:12:00 to say, does this actually work? They do like experiments. And then some combination of
02:12:07 neuroscience, where you’re starting to actually be able to observe, do certain experiments on
02:12:13 human beings and observe how the brain is actually functioning. And there, using intuition, you can
02:12:21 start being the philosopher. Like Richard Feynman is the philosopher, cognitive psychologists can
02:12:26 become the philosopher, and psychiatrists become much more like doctors. They’re like very medical.
02:12:32 They help people with medication, biochemistry, and so on. But they are no longer the book writers
02:12:39 and the philosophers, which of course I admire. I admire the Richard Feynman ability to do
02:12:45 great low level mathematics and physics and the high level philosophy.
02:12:52 Yeah, I think it was Fromm and Jung more than Freud that was sort of initially kind of like
02:13:00 made me feel like, oh, this is really amazing and interesting and I want to explore it further.
02:13:06 I actually, when I got to college and I lost that thread, I found more of it in sociology
02:13:15 and literature than I did in any place else. So I took quite a lot of both of those
02:13:23 disciplines as an undergraduate. And I was actually deeply ambivalent about
02:13:32 the psychology because I was doing experiments after the initial flurry of interest in
02:13:40 why people would occupy buildings during an insurrection and consider
02:13:44 being so overcommitted to their beliefs. But I ended up in the psychology laboratory running
02:13:55 experiments on pigeons. And so I had these profound dissonance between the kinds of issues
02:14:03 that would be explored when I was thinking about what I read about in modern British literature
02:14:12 versus what I could study with my pigeons in the laboratory. That got resolved when I went
02:14:18 to graduate school and I discovered cognitive psychology. And so for me, that was the path
02:14:25 out of this sort of like extremely sort of ambivalent divergence between the interest
02:14:31 in the human condition and the desire to do actual mechanistically oriented thinking about it. And I
02:14:42 think we’ve come a long way in that regard and that you’re absolutely right that nowadays this
02:14:50 is something that’s accessible to people through the pathway in through computer science or the
02:14:57 pathway in through neuroscience. You can get derailed in neuroscience down to the bottom of
02:15:08 the system where you might find the cures of various conditions, but you don’t get a chance
02:15:16 to think about the higher level stuff. So it’s in the systems and cognitive neuroscience and
02:15:21 computational intelligence, miasma up there at the top that I think these opportunities are most
02:15:28 are richest right now. And so yes, I am indeed blessed by having had the opportunity to fall
02:15:36 into that space. So you mentioned the human condition, speaking which you happen to be a
02:15:44 human being who’s unfortunately not immortal. That seems to be a fundamental part of the human
02:15:52 condition that this ride ends. Do you think about the fact that you’re going to die one day? Are you
02:16:00 afraid of death? I would say that I am not as much afraid of death as I am of degeneration. And
02:16:15 I say that in part for reasons of having, you know, seen some tragic degenerative situations
02:16:24 unfold. It’s exciting when you can continue to participate and feel like you’re near the place
02:16:42 where the wave is breaking on the shore, if you like. And I think about my own future potential.
02:16:58 If I were to begin to suffer from Alzheimer’s disease or semantic dementia or some other
02:17:07 condition, you know, I would sort of gradually lose the thread of that ability. And so one can
02:17:17 live on for a decade after, you know, sort of having to retire because one no longer has
02:17:28 these kinds of abilities to engage. And I think that’s the thing that I fear the most.
02:17:34 SL. The losing of that, like the breaking of the wave, the flourishing of the mind,
02:17:42 where you have these ideas and they’re swimming around and you’re able to play with them.
02:17:46 RL. Yeah. And collaborate with other people who, you know, are themselves
02:17:54 really helping to push these ideas forward. So, yeah.
02:17:58 SL. What about the edge of the cliff? The end? I mean, the mystery of it. I mean…
02:18:05 RL. The migrated sort of conception of mind and, you know, sort of continuous sort of way of
02:18:12 thinking about most things makes it so that, to me, the discreteness of that transition is less
02:18:25 apparent than it seems to be to most people.
02:18:27 SL. I see. I see. Yeah. Yeah. I wonder, so I don’t know if you know the work of Ernest Becker
02:18:35 and so on. I wonder what role mortality and our ability to be cognizant of it
02:18:42 and anticipate it and perhaps be afraid of it, what role that plays in our reasoning of the world.
02:18:49 RL. I think that it can be motivating to people to think they have a limited period left.
02:18:55 SL. I think in my own case, you know, it’s like seven or eight years ago now that I was
02:19:03 sitting around doing experiments on decision making that were
02:19:11 satisfying in a certain way because I could really get closure on whether the model fit the data
02:19:19 perfectly or not. And I could see how one could test, you know, the predictions in monkeys as well
02:19:26 as humans and really see what the neurons were doing. But I just realized, hey, wait a minute,
02:19:33 you know, I may only have about 10 or 15 years left here. And I don’t feel like I’m getting
02:19:40 towards the answers to the really interesting questions while I’m doing this particular level
02:19:46 of work. And that’s when I said to myself, okay, let’s pick something that’s hard. So that’s when
02:19:56 I started working on mathematical cognition. And I think it was more in terms of, well,
02:20:03 I got 15 more years possibly of useful life left. Let’s imagine that it’s only 10.
02:20:09 I’m actually getting close to the end of that now, maybe three or four more years.
02:20:13 But I’m beginning to feel like, well, I probably have another five after that. So, okay, I’ll give
02:20:17 myself another six or eight. But a deadline is looming and therefore. It’s not going to go on
02:20:23 forever. And so, yeah, I got to keep thinking about the questions that I think are the interesting and
02:20:31 important ones for sure. What do you hope your legacy is? You’ve done some incredible work in
02:20:37 your life as a man, as a scientist, when the aliens and the human civilization is long gone
02:20:46 and the aliens are reading the encyclopedia about the human species. What do you hope is the
02:20:51 paragraph written about you? I would want it to sort of highlight
02:20:56 a couple things that I was able to see one path that was more exciting to me than the one that
02:21:20 seemed already to be there for a cognitive psychologist, but not for any super special
02:21:28 reason other than that I’d had the right context prior to that, but that I had gone ahead and
02:21:34 followed that lead. And then I forget the exact wording, but I said in this preface that
02:21:44 the joy of science is the moment in which a partially formed thought in the mind of one person
02:22:01 gets crystallized a little better in the discourse and becomes the foundation
02:22:08 of some exciting concrete piece of actual scientific progress. And I feel like that
02:22:16 moment happened when Rumelhart and I were doing the interactive activation model and when
02:22:21 Rumelhart heard Hinton talk about gradient descent and having the objective function to guide the
02:22:29 learning process. And it happened a lot in that period and I sort of seek that kind of
02:22:37 thing in my collaborations with my students. So the idea that this is a person who contributed
02:22:49 to science by finding exciting collaborative opportunities to engage with other people
02:22:55 through is something that I certainly hope is part of the paragraph.
02:22:59 And like you said, taking a step maybe in directions that are non obvious. So it’s the
02:23:08 old Robert Frost road less taken. So maybe because you said like this incomplete initial idea,
02:23:16 that step you take is a little bit off the beaten path.
02:23:22 If I could just say one more thing here. This was something that really contributed
02:23:28 to energizing me in a way that I feel it would be useful to share. My PhD dissertation project
02:23:40 was completely empirical experimental project. And I wrote a paper based on the two main
02:23:48 experiments that were the core of my dissertation and I submitted it to a journal. And at the end
02:23:55 of the paper, I had a little section where I laid out the beginnings of my theory about what I
02:24:05 thought was going on that would explain the data that I had collected. And I had submitted the
02:24:13 paper to the Journal of Experimental Psychology. So I got back a letter from the editor saying,
02:24:20 thank you very much. These are great experiments and we’d love to publish them in the journal.
02:24:23 But what we’d like you to do is to leave the theorizing to the theorists and take that part
02:24:30 out of the paper. And so I did, I took that part out of the paper. But I almost found myself labeled
02:24:42 as a non theorist by this. And I could have succumbed to that and said, okay, well, I guess
02:24:50 my job is to just go on and do experiments, right? But that’s not what I wanted to do. And so when I
02:25:01 got to my assistant professorship, although I continued to do experiments because I knew I had
02:25:07 to get some papers out, I also at the end of my first year submitted my first article to
02:25:13 Psychological Review, which was the theoretical journal where I took that section and elaborated
02:25:18 it and wrote it up and submitted it to them. And they didn’t accept that either, but they said,
02:25:24 oh, this is interesting. You should keep thinking about it this time. And then that was what got me
02:25:29 going to think, okay, you know, so it’s not a superhuman thing to contribute to the development
02:25:37 of theory. You know, you don’t have to be, you can do it as a mere mortal.
02:25:43 LB And the broader, I think, lesson is don’t succumb to the labels of a particular reviewer.
02:25:50 RL Yeah, that’s for sure. Or anybody labeling you, right?
02:25:55 LB Yeah, exactly. I mean that, yeah, exactly. And especially as you become successful,
02:26:01 your labels get assigned to you for that you’re successful for that thing.
02:26:05 RL Connectionist or cognitive scientist and not a neuroscientist.
02:26:09 LB And then you can, you can completely, that’s just, that’s the stories of the past. You’re
02:26:15 today a new person that can completely revolutionize in totally new areas. So don’t
02:26:20 let those labels hold you back. Well, let me ask the big question. When you look at into the,
02:26:29 you said it started with Columbia trying to observe these humans and they’re doing
02:26:34 weird stuff and you want to know why are they doing this stuff. So Zuma even bigger.
02:26:38 LB At the hundred plus billion people who’ve ever lived on earth. Why do you think we’re all
02:26:47 doing what we’re doing? What do you think is the meaning of it all? The big why question.
02:26:51 We seem to be very busy doing a bunch of stuff and we seem to be kind of directed towards somewhere.
02:26:59 But why?
02:27:00 RL Well, I myself think that we make meaning for ourselves and that we find inspiration
02:27:13 in the meaning that other people have made in the past. You know, and the great religious thinkers
02:27:21 of the first millennium BC and, you know, few that came in the early part of the second millennium,
02:27:36 you know, laid down some important foundations for us.
02:27:40 But I do believe that, you know, we are an emergent result of a process that happened
02:27:54 naturally without guidance and that meaning is what we make of it and that the creation of
02:28:05 efforts to reify meaning in like religious traditions and so on is just a part of the
02:28:15 expression of that goal that we have to, you know, not find out what the meaning is, but to
02:28:26 make it ourselves. And so, to me, it’s something that’s very personal. It’s very individual. It’s
02:28:40 like meaning will come for you through the particular combination of synergistic elements
02:28:50 that are your fabric and your experience and your context and, you know, you should…
02:29:04 It’s all made in a certain kind of a local context though, right? Here I am at UCSD with this brilliant
02:29:12 man, Rommelhart, who’s having, you know, these doubts about symbolic artificial intelligence
02:29:24 that resonate with my desire to see it grounded in the biology and let’s make the most of that,
02:29:35 you know? Yeah. And so, from that like little pocket, there’s some kind of peculiar little
02:29:41 emergent process that then, which is basically each one of us, each one of us humans is a kind of,
02:29:49 you know, you think cells and they come together and it’s an emergent process that then tells fancy
02:29:56 stories about itself and then gets, just like you said, just enjoys the beauty of the stories
02:30:03 we tell about ourselves. It’s an emergent process that lives for a time, is defined by its local
02:30:10 pocket and context in time and space and then tells pretty stories and we write those stories
02:30:16 down and then we celebrate how nice the stories are and then it continues because we build stories
02:30:21 on top of each other and eventually we’ll colonize hopefully other planets, other solar systems,
02:30:30 other galaxies and we’ll tell even better stories. But it all starts here on Earth. Jay, you’re
02:30:37 speaking of peculiar emergent processes that lived one heck of a story. You’re one of the
02:30:47 the great scientists of cognitive science, of psychology, of computation. It’s a huge honor
02:30:58 you would talk to me today that you spend your very valuable time. I really enjoyed talking with
02:31:03 you and thank you for all the work you’ve done. I can’t wait to see what you do next.
02:31:06 JL Well, thank you so much and this has been an amazing opportunity for me to let ideas that I’ve
02:31:13 never fully expressed before come out because you asked such a wide range of the deeper questions
02:31:20 that we’ve all been thinking about for so long. So thank you very much for that.
02:31:24 RL Thank you. Thanks for listening to this conversation with Jay McClelland.
02:31:29 To support this podcast, please check out our sponsors in the description.
02:31:32 And now, let me leave you with some words from Jeffrey Hinton. In the long run,
02:31:37 curiosity driven research works best. Real breakthroughs come from people focusing
02:31:43 on what they’re excited about. Thanks for listening and hope to see you next time.